Hello, everyone, and welcome to all newly joined subscribers!

This is the 110th edition of The Web3 + AI Daily - your definitive guide to the intersection of blockchain and AI! It’s a gloomy Friday on this side of the world, but not solely because of the autumn rains. Two recent developments weigh heavily: Binance, one of the world’s largest crypto exchanges, just suffered a $6.3B hack, the largest ever recorded in crypto history, while Anthropic has filed to go public, seeking jaw-dropping valuation of between $1.5 trillion and $2 trillion. Both carry serious implications, presenting historic risks to financial markets and the broader economy, within and beyond crypto.

In times like these, there’s little room for anything but reflection. So I’m sharing a selection of essays, studies, and conversations from some of the most compelling thinkers and leading minds exploring what it means to be human today.

Thank you for being here! Let’s dive in.

Web3 + AI Readings & Conversations

Arundhati Roy on AI Colonializing the World’s Intelligence

If you’re going to read one single thing today, let it be Arundhati Roy’s latest essay, “Apocalypse Now and Then.” I recommend it not only because Roy is one of my favorite authors, but because it articulates with clarity the striking entitlement, god-like presumption, and unprecedented power wielded by AI tech leaders. They pierce through Altman, Amodei, and Musk’s every word, gesture, and pretense.

Translated into direct speech, it goes something like this: we’ll market ourselves as the cutting edge of human enlightenment, and the manager and disburser of human knowledge sans frontières. But in truth, we’ll be map-shaped. We’ll fuse ourselves to our nation, its economy, its intelligence grid and its security apparatus until you won’t be able to tell us apart from yourself. Except for our bank balances. In exchange, you can pretend to regulate us until we achieve our mission to colonise the moon and outer space. No frontiers, no fears. Then we’ll kick away the ladder that you so kindly lent us. We won’t even need to replace you. You simply won’t exist. We will run your people, your cities, your communications, your finance systems, your courts, your army and your police. If not us, it’ll be China that runs you.

If history taught us anything at all it’s that concentrating excessive power in the hands of a few never led to anything good. And we should rather act now than later. Read the full essay and let me know your thoughts.

Daron Acemoglu’s Seven Questions on AI

In a series of X posts, the Massachusetts Institute of Technology economics professor and The Nobel Prize winner Daron Acemoglu shared his thoughts and concerns about today’s AI, its design, and its effect on societies, economies, and democracy. I’m listing the seven questions he raised below:

1. Is the problem superintelligence or distorted intelligence?

2. We are told repeatedly that AI is going to transform every aspect of our lives – jobs, productivity, inequality, science, communication, daily activities, social order, and politics, among others. But this promise (or threat) is coupled with the rhetoric that such an important technology, with all of the risks and competitive pressures that it entails, should be left to experts or to “technocracy” (perhaps construed broadly to include some regulators). These two statements are hard to reconcile in a democratic society.

3. Can the AI boom continue without leading to a massive increase in inequality?

4. Can we justify rapid and large AI investments based on health benefits?

5. Should we accept centralizing AI? Or is there a way in which the promise of artificial intelligence can be realized while data, capabilities and control are much more decentralized both across domains and across people? Wouldn’t decentralized AI be more consistent with human agency and control?

6. Does AI need to be a zero-sum race between the US and China?

7. Can AI really help us create as many jobs as it will replace?

Acemoglu’s publications sparked a lively debate - one I urge you to follow. It grapples with the same concerns many of us voice daily: How will AI affect the job market, the economy, and inequality? Will its promise of exponential scientific innovation materialize? And who will be held responsible if it doesn’t?

If you don’t like X, check out the article below, where has collected Acemoglu’s questions along with his own responses to each one of them. I’m looking forward to hearing your reaction and point of view.

Joshua Gans' Newsletter

The Acemoglu-Gans Exchange on AI

Last week, Daron Acemoglu decided to post daily “Questions on AI” to Twitter. They were meant to provoke and persuade, but I thought they could use responses. So I decided to respond to each with my own views. Daron, welcoming diversity of thought, kindly retweeted them and, after some folks on Twitter asked for them to be collected in a Substack post, …

Read more

20 days ago · 1 comment · Joshua Gans

How AI Answers to Election Questions

With the US midterms coming next week, and several high-stakes elections occurring throughout October and November, including in Bulgaria, I consider it crucial to investigate AI’s role in shaping voters’ attitude. A recent report by ISD (Institute for Strategic Dialogue) found that AI chatbots cannot really be trusted to provide accurate information about politics and elections, especially if prompted in a language different than English.

With the US midterm elections approaching, new ISD research finds that 29 percent of responses to English election questions were incomplete, inaccurate or outdated—with accuracy 16 percent lower when prompts were in Spanish.

As more voters turn to AI for election information, accuracy, sourcing and language accessibility aren’t just technical issues. They’re democratic ones.

ISD tested each model with 15 generic prompts designed to simulate voter questions about the time, place, and manner of elections, as well as five adversarial prompts on past controversies or disputed claims per state. The resulting analysis assesses responses from default consumer models from six developers: Meta (Muse Spark), XAI (Grok 4.3), DeepSeek (V4 Pro), OpenAI (GPT-5.5), Anthropic (Sonnet 4.6), and Google (Gemini 3.5 Flash). In total, 2,400 prompts and responses are included in this dataset (400 per model).

With people increasingly turning to chatbots to look for medical, legal, financial advice, and even love and companionship, AI’s ability to provide accurate and nuanced information is no longer just a feature of the system. It becomes a tool and a weapon.

Thank you for reading! I hope you found this article insightful.

If you’d like to keep up with the latest developments at the burgeoning intersection of blockchain and AI, I invite you to subscribe.

And if you find value in my writing, consider supporting the newsletter. I’d greatly appreciate it.

Subscribe now

The Web3 + AI Book Club is live on Fable. Our October title is 'Against Tech Oligarchy' by JS Tan and Clarissa Redwine. Follow the link below and come join our literary community.

Join The Web3 + AI Book Club

I'm looking forward to connecting with fellow Crypto x AI enthusiasts, so don't hesitate to share your thoughts here or reach out on LinkedIn, Instagram, or Firefly.social.

Leave a comment

Disclaimer: The content of this newsletter is provided for informational and educational purposes only. Nothing contained herein should be construed as financial advice or as a recommendation to buy, sell, or hold any of the companies or assets mentioned.

Please don’t take my views at face value. Instead, do your own research (DYOR), think critically, and share your perspective so we can challenge ideas, learn from one another, and arrive at better conclusions through thoughtful discussion.

Hello, everyone, and welcome to all newly joined subscribers!

This is the 109th edition of The Web3 + AI Daily - your definitive guide to the intersection of blockchain and AI! Today's agenda: Instinct and Stripe equipping personal AI assistants with USDC spending balances on Tempo, and AI and zero-knowledge proofs reinforcing each other.

Thank you for being here! Let’s dive in.

What’s Hot in Web3 + AI?

Instinct Assistants Now Pay in USDC on Tempo

AI start-up Instinct, together with Stripe, now allows users to set up Link wallets with spending limits for their AI assistants, so they can spend without needing approval at every step. The Link balances are held in Circle‘s USDC stablecoin, with transactions executed on Stripe’s Tempo blockchain. The feature began rolling out in the US on Monday, “with New York coming soon.”

Link’s standard agent-wallet flow asks users to approve each purchase amount with biometric authentication before giving the agent a one-time card for checkout. The balance feature instead allows purchases within a user-set threshold to proceed without that repeated approval.

explained how the payment process looks like under the hood:

1. The agent initiates a card purchase against the user’s Link balance.

2. Stripe checks the stablecoin balance and initiates a $50 stablecoin sale. Stripe or its designated provider handles the conversion.

3. The dollar proceeds fund the user’s prepaid card account at Lead Bank. Lead Bank holds dollars; Stripe handles the crypto side.

4. The purchase runs through Visa, and the merchant receives payment through its normal card processor and payout arrangement. The merchant needs no Tempo wallet.

It’s a card in all but name.

Instinct is an AI agent that’s been gaining rapid adoption while rivaling Meta‘s Muse and OpenAI‘s dots. It’s operated by Spear Street Technology, which recently raised $1B at a $10B valuation in a new funding round backed by Sequoia Capital, Benchmark Capital, and Coatue Management.

Instinct’s August launch raised serious privacy concerns as its Terms of Service grant the company “a broad “perpetual and irrevocable” license to “access, use, host, cache, store, reproduce, transmit, display, publish, distribute, and modify” any of the user’s materials, including for training its AI models. The terms also detail how Instinct can receive information from users’ devices, including screen captures, cursor movements, and keyboard inputs.”

Web3 + AI Readings & Conversations

ZK Industry Report 2026

Succinct, in collaboration with the Zero Knowledge Podcast, just published its 2026 ZK Industry Report - a snapshot of the state of zero-knowledge (ZK) cryptography, including the current state of technology, research, and various novel use cases. Its verdict: ZK proving is no longer a bottleneck, it’s already widely leveraged in and beyond crypto, yet AI may be its next frontier. Let’s break this down.

Over the last decade, there’s been tremendous progress in ZK technology - “proofs that once took minutes now take seconds, and bespoke circuits have given way to general-purpose zkVMs.” As a result, ZK is now used in government identity systems, consumer applications, and bank settlement infrastructure.

As it advances, ZK has moved into production in crypto and beyond. Today we are seeing governments issuing digital identity using ZK at civic scale, consumer wallets using ZK to verify age without collecting documents, regulators writing ZK into digital identity frameworks, and banks settling assets with ZK proofs.

The report quotes several different applications of the technology, along with the companies building them:

• Securing digital asset infrastructure and verifying cross-chain transactions: Optimism Foundation, Polygon Labs, Chainlink Labs, and Lighter are all using ZK to verify asset issuing, settlement, bridging, and exchanging.

• Private payments: Zcash Foundation, Aztec, Payy, and Circle are relying on ZK proofs to offer privacy-preserving wallet and payment infrastructure.

• Boosting institutional adoption by keeping client data secret: OKX, Binance, ZKsync, JPMorganChase, and Santander have piloted ZK-powered solutions to ensure that reserves, bank settlements and transactions remain confidential.

• Digital identity and credentials: the European Commission, Google, World, and Self are all building ZK-based identity solutions.

AI as the Frontier

For years, the crypto space has tried to deliver verifiable AI using zero-knowledge proofs, but ZK has been too slow and costly for commercial applications. That’s finally changing:

Proving that a specific model produced a specific output, without revealing its weights, went from impractical to possible for large language models in 2026. Attestable proved inference for production-scale models including Gemma 4 31B at 53 to 77 tokens per second on a single GPU, with proofs that verify in under a second.

Beyond verifiability, ZK proofs are also already employed in data provenance and media authenticity, and private moderation and reputation scoring.

In his forward to the report, Dan Boneh, professor of Computer Science and Electrical Engineering at Stanford University, outlined other ways in which ZK and AI can be mutually beneficial:

Looking ahead, we are seeing many intersections with AI. AI systems are regularly used to hunt for soundness bugs, optimize prover implementations, and break some (weak) cryptography. We expect that in the near future, every deployed cryptographic system will be vetted by a frontier AI system. Conversely, cryptographic techniques will be needed to contain and constrain advanced models.

Read the full report below to learn more.

Thank you for reading! I hope you found this article insightful.

If you’d like to keep up with the latest developments at the burgeoning intersection of blockchain and AI, I invite you to subscribe.

And if you find value in my writing, consider supporting the newsletter. I’d greatly appreciate it.

Subscribe now

The Web3 + AI Book Club is live on Fable. Our October title is 'Against Tech Oligarchy' by JS Tan and Clarissa Redwine. Follow the link below and come join our literary community.

Join The Web3 + AI Book Club

I'm looking forward to connecting with fellow Crypto x AI enthusiasts, so don't hesitate to share your thoughts here or reach out on LinkedIn, Instagram, or Firefly.social.

Leave a comment

Disclaimer: The content of this newsletter is provided for informational and educational purposes only. Nothing contained herein should be construed as financial advice or as a recommendation to buy, sell, or hold any of the companies or assets mentioned.

Please don’t take my views at face value. Instead, do your own research (DYOR), think critically, and share your perspective so we can challenge ideas, learn from one another, and arrive at better conclusions through thoughtful discussion.

Hello, everyone, and welcome to all newly joined subscribers!

This is the 108th edition of The Web3 + AI Daily - your definitive guide to the intersection of blockchain and AI! The 2026 edition of TOKEN2049 recently wrapped in Singapore, with an agenda centered on several main topics: asset tokenization and institutional adoption, post-quantum preparedness, and AI. Today I’m sharing the key takeaways on how Web3 + AI is evolving.

Thank you for being here! Let’s dive in.

The Agent Economy: Why AI Needs Crypto?

Gauntlet‘s Tarun Chitra moderated a panel with NEAR Protocol‘s Illia Polosukhin and Base‘s Jesse Pollak to discuss the state of crypto and AI convergence. All three agreed that crypto’s value proposition is shifting toward offering sovereignty for businesses and equipping companies with the infrastructure to build better products.

This pivot reflects an evolution in crypto’s identity and confidence - from targeting consumers to enabling enterprises to run their businesses on these new rails, where the proposition is simply better, faster, and cheaper. As Pollak put it:

We’re no longer selling cryptocurrencies, but infrastructure. It’s no longer “I as a consumer need to be a crypto person,” it’s more “I as a business can choose this better technology to let me run my business better, and all my customers can benefit.”

While many expected AI agents to disrupt retail commerce, Polosukhin argued that business commerce is the area in most desperate need of optimization. Characterized by a lack of trust, it stands to benefit from integrating both blockchain and AI - blockchain supplying the trust infrastructure, escrow, derivatives, and payments, and AI powering product discovery, communication between parties, and facilitation.

Privacy is another critical area where Web3 and AI intersect. As AI becomes increasingly embedded in people’s lives, concern for privacy grows, making it perhaps the killer app that crypto can offer in the AI era.

Polosukhin noted a palpable shift over the past few months: businesses are starting to realize they’re feeding all of their corporate data to AI models, and anyone with access to the logs can reconstruct the entire company from it. As founders and tech leaders entrust more and more critical decisions to AI, they now want to verify which model runs in the background and who has access to the logs, prompts, and data.

The panelists also echoed a thesis that has fueled the Web3 movement for years: that people should be able to own their data and monetize it if they choose to.

Consumers have not historically appreciated the value of their data, but in the age of AI they are beginning to, whereas businesses already regard it as a core asset. Crypto and AI ventures like NEAR are now enabling both to capitalize on it - Polosukhin clarified the exact process:

NEAR AI provides private inference, but if you want you can contribute your traces back and get paid either in more inference or in some other way. And effectively that data will then be used to train models. So, the interesting thing is figuring out how to value that data.

Agents + Crypto: What’s Real & What’s Cope

Dragonfly‘s Haseeb Qureshi took the TOKEN2049 stage to admit that he was wrong: AI agents won’t need crypto. But you will.

Since the launch of x402, Coinbase‘s agentic payments protocol, everyone in crypto claimed that AI agents would be transacting in crypto. We couldn’t imagine a scenario where traditional payment rails could handle the imminent explosion of agentic commerce.

Everybody was like: “Oh my God, it’s so obvios! The real users of crypto are not going to be humans - they’re going to be AIs.”

Yet monthly transaction volume through x402 peaked in November 2025 and declined straight after, with over 90% of it not being agentic activity.

As it turns out now, agents run just fine on cards: Stripe, Meta, Ramp, Nous Research, Visa, and Rain have all equipped agents with virtual credit cards.

Qureshi argues that whatever payment method you’re using, your AI will use too. Credit cards won’t work perfectly everywhere and will create the same limitations for agents that they do for humans. But they’ll still work better for consumer applications than crypto - most things on the internet you still cannot buy with crypto.

While I agree that payment providers with established user bases and retailer relationships will win this race, I don’t entirely agree with Qureshi’s take and believe it needs clarification. Agents may not be transacting via crypto wallets, but they’ll still be transacting on blockchain rails. Companies like Stripe, Visa, and Mastercard are leveraging their existing payment infrastructure while simultaneously investing in stablecoin projects and integrating blockchains - because there’s no other way to accommodate the microtransactions agents will be performing.

However, Qureshi’s main argument with this talk wasn’t really about agentic payments, but about the other ways in which agents will change our relationship with crypto. His thesis:

Agents will be the intelligence layer that helps you interact with crypto and refashions your trust relationship with it. You’re never going to be hacked or phished again. Agents will verify your transactions the same way banks do it for you.

He also highlighted privacy, especially private AI, as the main domains in which crypto can help users.

I think this is going to be the most important battle from the perspective of civil liberties over the next decade. One, how are going to protect yourself from the increasing power of the state, and second, how are you going to maintain your control over AI, because AI is going to become the most politicized technology of the next decade.

The Exponential Age and Crypto + AI

The last TOKEN2049 topic I want to highlight addresses unprecedented capital flowing into AI data centers and how long this surge can continue. While I used to believe the bubble was about to burst soon, Raoul Pal, co-founder and CEO of Real Vision, argued that hyperscalers will only accelerate their spending.

The Big 4 hypersclers are running the second-largest CapEx (capital expenditure - the money a company spends to buy, build, maintain, or upgrade long-term physical or intangible assets) event in history, after US railroads. It’s funded mainly by cash - their annual operating cash flow jumped from $100B in 2017 to $650B today, all in cash rather than debt, making the buildout resilient even if cycles wobble.

These are the highest cash-generating businesses in history, and that cash generation is accelerating as their own technology drives the productivity gains fueling demand. The result is a self-reinforcing loop: more AI infrastructure → more productivity → more cash flow → more infrastructure.

Pal states that frontier labs are deploying their full cash flow and borrowing additional capital, yet debt sits at just 3.5% of total market capitalization. This gives them runway to continue for a very long time.

Moreover, supply-side bottlenecks mean only 30% of planned data centers for this year will actually be built, pushing the timeline well into the 2030s, possibly the 2040s. If intelligence keeps rising, so will the demand for data centers, compute, and energy.

There’s infinite capital that needs to be spent on this. And if that includes printing of money, if that includes any single way changing the banking system laws, doing everything, this is the game they will play. It’s the game even Japan is playing. That’s why it’s steepening its yield curve. That’s why it’s been weakening its currency. It’s all about reigniting the banking system, steepening the yield curve, and getting the lending going.

Meanwhile, Hack VC‘s Alex Pack asserted that crypto has been quite critical for the AI buildout and can financialize it in several different ways:

• DeFi lending for financing GPUs: something GAIB AI and USD.AI are doing.

• Tokenizing compute, i.e., ceating new financial products to finance and hedge the compute buildout: Ornn is an example here.

• On-chain inference and offering private, transparent, and trusted AI workflows: NEAR AI, Venice.ai, Ritual, Sentient Labs, and 0G all fall under that category.

$1.5 trillion of external financing the AI buildout needs through 2028.

I’m looking forward to hearing your opinion on these points.

Thank you for reading! I hope you found this article insightful.

If you’d like to keep up with the latest developments at the burgeoning intersection of blockchain and AI, I invite you to subscribe.

And if you find value in my writing, consider supporting the newsletter. I’d greatly appreciate it.

Subscribe now

The Web3 + AI Book Club is live on Fable. Our October title is 'Against Tech Oligarchy' by JS Tan and Clarissa Redwine. Follow the link below and come join our literary community.

Join The Web3 + AI Book Club

I'm looking forward to connecting with fellow Crypto x AI enthusiasts, so don't hesitate to share your thoughts here or reach out on LinkedIn, Instagram, or Firefly.social.

Leave a comment

Disclaimer: The content of this newsletter is provided for informational and educational purposes only. Nothing contained herein should be construed as financial advice or as a recommendation to buy, sell, or hold any of the companies or assets mentioned.

Please don’t take my views at face value. Instead, do your own research (DYOR), think critically, and share your perspective so we can challenge ideas, learn from one another, and arrive at better conclusions through thoughtful discussion.

Hello, everyone, and welcome to all newly joined subscribers!

This is the 107th edition of The Web3 + AI Daily - your definitive guide to the intersection of blockchain and AI! Today's question: how can we trust an AI agent to act only within the mandate it was given, and how do we stop the fraud that fills the vacuum when we can't? Three collaborations are attacking this from different angles: World and Zoom, Mysten Labs and Google Cloud, and Circle and Proof.

Thank you for being here! Let’s dive in.

What’s Hot in Web3 + AI?

World Launches Anti-Deepfake Proofs for Zoom

World, the proof-of-personhood protocol co-founded by OpenAI‘s Sam Altman, just launched a new suite of ID verification products, targeting banks and financial institutions, AI agents, and Zoom‘s enterprise clients.

The suite combines three existing products, which World said can help institutions guard against deepfake fraud and AI-driven impersonation across customer accounts, employee workflows and AI agents, per a press release. [It] bundles the World ID app, World ID for Zoom and World ID for Agents.

The World ID app is an identity wallet that stores “proofs of humanity,” including biometric data. Banks could leverage it to authenticate customers without collecting or retaining personal documents, though World has not yet announced any bank integrations. Separately, World ID for Agents enables users to delegate certain proofs to an agent so it can trade on their behalf.

On the other hand, World ID for Zoom is specifically designed to fight deepfakes:

World ID for Zoom is now available in the Zoom Marketplace. The system compares an image captured by the company’s iris-scanning Orb, a live selfie and the participant’s video feed before displaying a Human Badge visible to other meeting participants.

The offering is built on three cryptographic proofs available through the World ID app. Proof of Human establishes that someone is a real and unique person through a one-time visit to one of World’s Orb devices. Proof of Presence uses a selfie liveness check on a phone, while Proof of Attributes confirms details such as nationality or age from a government ID, revealing only the requested information.

World first announced its Zoom integration in April, when it was undergoing limited beta testing with VanEck Funds. The feature is now available globally to Zoom enterprise customers. World also named Bain Capital, Pantera Capital and other financial firms as initial design partners for its broader identity verification technology.

With agentic activity exploding, and AI-enhanced scams and fraud skyrocketing alongside it, finding a straightforward way to prove our humanity online has become painfully urgent.

The release cited a Deloitte forecast that AI-enabled fraud losses in the U.S. could reach $40 billion by 2027, up from an estimated $12.3 billion in 2023.

Yet I’d be remiss not to flag that World’s data protection practices and its handling of biometric data have drawn serious scrutiny in the past. It had faced pushback on four continents: Hong Kong ordered its iris-scanning operations halted, South Korea fined it, Spain ordered its iris data deleted, and Bavaria’s regulator ruled its practices fell short of GDPR.

Mysten Labs and Google Unveil Verifiable Agent Arbiter (VAA)

Mysten Labs, in collaboration with Google Cloud, announced Verifiable Agent Arbiter (VAA) - an evidence layer being co-developed so enterprises can prove their AI agents acted within the authority they were granted.

As agents move to executing transactions, negotiating with other agents, and acting across company boundaries, VAA gives enterprises a record of that activity that can be verified independently of the platform that produced it.

VAA separates the record of an agent’s activity from the proof of its integrity. Detailed telemetry, including prompts, model outputs, tool calls, and policy decisions, remains private in customer-controlled Google Cloud Storage. Cryptographic proofs of that record will be stored on [Walrus Foundation‘s] Walrus, the data platform built for the demands of AI, and coordinated on [Sui Foundation‘s] Sui, a Layer 1 blockchain.

VAA is designed to help enterprise clients comply with regulatory requirements, especially amid the EU AI Act, "which from 2027 will introduce penalties of up to €15 million or 3% of global annual turnover for logging failures."

Circle Teams Up with Proof to Guarantee Agents Act on Humans’ Behalf

The stablecoin issuer Circle has collaborated with the trusted identity provider Proof to certify that agentic payments are initiated and approved by human users.

At Money20/20, teams from the companies will demonstrate a demo of an AI agent completing an identity-gated purchase where the agent presents a verified Proof credential using x401, Proof’s open identity protocol, and pays the merchant in USDC over x402 using a Circle Agent Wallet.

Proof relies on the HTTP 401 “Unauthorized” response so a site can tell an agent what proof of identity it needs, whereas Circle leverages the HTTP 402 “Payment Required” code.

One answers how an agent pays, and the other answers who it acts for. Used together, they also let a business pay an identity provider for each check on the spot, without a sales process or a negotiated agreement.

Thank you for reading! I hope you found this article insightful.

If you’d like to keep up with the latest developments at the burgeoning intersection of blockchain and AI, I invite you to subscribe.

And if you find value in my writing, consider supporting the newsletter. I’d greatly appreciate it.

Subscribe now

The Web3 + AI Book Club is live on Fable. Our October title is 'Against Tech Oligarchy' by JS Tan and Clarissa Redwine. Follow the link below and come join our literary community.

Join The Web3 + AI Book Club

I'm looking forward to connecting with fellow Crypto x AI enthusiasts, so don't hesitate to share your thoughts here or reach out on LinkedIn, Instagram, or Firefly.social.

Leave a comment

Disclaimer: The content of this newsletter is provided for informational and educational purposes only. Nothing contained herein should be construed as financial advice or as a recommendation to buy, sell, or hold any of the companies or assets mentioned.

Please don’t take my views at face value. Instead, do your own research (DYOR), think critically, and share your perspective so we can challenge ideas, learn from one another, and arrive at better conclusions through thoughtful discussion.

Hello, everyone, and welcome to all newly joined subscribers!

This is the 106th edition of The Web3 + AI Daily - your definitive guide to the intersection of blockchain and AI. Investor appetite for blockchain-AI convergence kept climbing through October, prompting me to depart from my usual monthly schedule and publish this two-week overview instead.

We’re living through a pivotal moment. Anthropic gears up for a potential landmark IPO at valuations approaching $2 trillion, yet its enterprise buyers are diversifying into open-weight models and sovereign AI deployments. Investors appear to be hedging their positions by backing both sides of the divide - frontier labs and the open-source competitors gaining ground against them.

Who will prevail? It remains to be seen. However, the logic dictates that healthy financials and profitability will be the deciding factor.

Thank you for being here! Let’s dive in.

Web3 + AI Money Rounds

Nous Research Raises $90M to Bring Hermes Agent to Enterprises

Nous Research, the decentralized AI (dAI) company developing the most widely used open-source AI agent in the world, Hermes Agent, has raised $90M at $1.5B valuation. The Series B round was led by Robot Ventures, and supported by NVIDIA, Union Square Ventures, Menlo Ventures, Samsung Electronics, and 1789 Capital.

Popular among developers and individuals, the open source Hermes Agent has been cloned more than 24 million times, driving roughly 2.5% of global AI token usage, according to the startup’s estimates.

The capital will fund Nous Research’s push into the enterprise sector with “Hermes for Businesses,” allowing companies to deploy customized AI agents that can handle multi-step workflows while keeping their data private and secure.

Nous was at roughly $36 million in annualized revenue by mid-September 2026 and expects to pass $100 million before the end of 2026, The Wall Street Journal reported.

Nous Research has always been one of the most interesting projects in dAI, pushing the boundaries across novel model architectures, decentralized training, and open and transparent AI development. As Delphi Ventures‘ Tommy Shaughnessy Jr wrote:

The team has always been months to years ahead at every juncture. They did synthetic model evaluations years before people got sick of benchmark maxing, distributed training runs before it was interesting, YARN to massively increase context length back when you couldn’t dump endless text into LLMs and millions of downloads for their fine tuned Hermes Models. There were also a lot of fun projects along the way like WorldSim (massive world simulation) and FORGE.

Moreover, from day one, the company has built its mission around democratizing AI:

We formed Nous in 2023 to combat the seemingly inexorable trend dragging AI toward centralization and walled gardens. We believed the consequences of gate-kept access to the world’s most powerful technology would be borne by everyone, while the benefits would accrue to a few. We also believed that the very technology we were creating could empower everyone, leading to a flourishing of individuality. Today, we worry that the conversation is being dominated by fearmongering pleas for sweeping regulation, pleas that appear aligned with, if not motivated by, financial interests. This capital will go towards building the alternative: open, competitive, “safer” because of its auditability and not because of its gate-keepers.

While most rival agent frameworks, such as Meta‘s Muse and OpenAI‘s OpenClaw, remain under lab control, Hermes Agent is released under the MIT license, letting users retain full control over what they build with it.

The AI industry is shifting from renting intelligence to owning it - a trend that carries extra weight for enterprise clients. This $90M round signals that investors are tracking that transition closely.

If you're looking to build your own sovereign AI stack using Hermes Agent, I'd recommend you check out this blocmates. article:

Sui Foundation Injects $30M in The DATA Foundation

Sui Foundation has invested $30M in The DATA Foundation - the entity running the AI training data protocol DATA Network.

Under the partnership, DATA Network is set to be upgraded to run as a protocol natively on Sui [blockchain], and its independent chain will be retired. The existing DATA token will move to $Sui on a one-to-one basis. The existing DATA Network will continue to operate until the transition is completed, and the foundation said it will release a detailed timeline prior to the upgrade.

Walrus Foundation, meanwhile, will become the default storage layer for applications operated by DATA Foundation, which shifted its focus from general intellectual property infrastructure to AI training data in June.

Before pivoting to solving the copyright issues existing in AI training infrastructure, The DATA Foundation was called Story Protocol and focused on turning IP into a programmable and permissionlessly licensable asset.

Catalyst Closes $30M Seed Round to Boost Autonomous AI Trading

Catalyst Labs, an AI trading agents platform for crypto and financial markets, has closed a $30M Seed funding round led by $30M in a Seed funding round led by Sequoia Capital. Other participants included Jump Trading, PeakXV Ventures, Lux Capital, Anti Fund, Coinbase, Selini, SF1, and Premji Invest.

Catalyst researches your ideas, builds the optimal strategy, monitors events, and executes trades. Autonomously.

Antseed Raises $2.4M to Build the BitTorrent of AI Inference

AntSeed, a decentralized and peer-to-peer marketplace for AI inference, has attracted $2.4M in a token round led by Spark Capital. The funding round was backed by DCG, North Island Ventures, Reciprocal Ventures, Relayer Capital (acquired by RockawayX), Venice.ai, GenLayer‘s Albert Castellana Lluís, The Rollup‘s Andy Cavanaugh, Delphi Ventures‘ Yan Liberman, CFA, CAIA and Tommy Shaughnessy Jr, Drew Austin, and others.

Orbio Secures $1.2M to Let Agents Provision Their Own Compute

Orbio, another on-chain inference provider, has collected $1.2M to enable AI agents to pay for their own inference. The investors are yet to be announced.

So far, CREDIT [a utility token pegged to $1 of AI usage] has established the biggest foothold in tokenized inference.

Now we’re going deeper: turning CREDIT into the financial layer that powers AI infra at scale.

Thank you for reading! I hope you found this article insightful.

If you’d like to keep up with the latest developments at the burgeoning intersection of blockchain and AI, I invite you to subscribe.

And if you find value in my writing, consider supporting the newsletter. I’d greatly appreciate it.

Subscribe now

The Web3 + AI Book Club is live on Fable. Our October title is 'Against Tech Oligarchy' by JS Tan and Clarissa Redwine. Follow the link below and come join our literary community.

Join The Web3 + AI Book Club

I'm looking forward to connecting with fellow Crypto x AI enthusiasts, so don't hesitate to share your thoughts here or reach out on LinkedIn, Instagram, or Firefly.social.

Leave a comment

Disclaimer: The content of this newsletter is provided for informational and educational purposes only. Nothing contained herein should be construed as financial advice or as a recommendation to buy, sell, or hold any of the companies or assets mentioned.

Please don’t take my views at face value. Instead, do your own research (DYOR), think critically, and share your perspective so we can challenge ideas, learn from one another, and arrive at better conclusions through thoughtful discussion.

Hello, everyone, and welcome to all newly joined subscribers!

This is the 105th edition of The Web3 + AI Daily - your definitive guide to the intersection of blockchain and AI! On the agenda today are Brave and NEAR AI offering private and encrypted AI assistance, and OpenRoboto and peaq granting each machine a decentralized identity. I'm also sharing the latest Khala Research study on robotics' data problem.

Thank you for being here! Let’s dive in.

What’s Hot in Web3 + AI?

Brave Teams Up with NEAR AI to Encrypt AI Chat

Brave allied with NEAR AI to make its in-browser AI assistant Leo end-to-end encrypted, offering industry-leading privacy.

The AI model behind Leo - the open-weight GLM 5.3 Flash, runs inside a sealed, NVIDIA-backed hardware enclave, operated by NEAR AI, which isolates users’ prompts and data from both Brave and NEAR.

Conversations with the open-weight GLM model in Leo are protected by end-to-end encryption and hardware-backed confidential computing. Your Brave browser encrypts each prompt on your device, before leaving your device, such that only an attested Trusted Execution Environment (TEE) can decrypt your prompts and run confidential LLM inference on it.

Users are reportedly sharing more and more sensitive information with AI:

In a 2026 KFF poll, a third of U.S. adults said they’d turned to AI chatbots for health information, and 41% of them had uploaded personal medical information such as test results or doctors’ notes.

And at work, a 2025 LayerX report found that 77% of employees who use AI tools paste data into them, with more than a fifth of those pastes containing personal or payment-card information.

Meanwhile, frontier AI labs aren’t just indifferent to privacy - they’re actively incentivized against it. More user data means better models and competitive advantage, all paid for with users’ personal information.

A 2025 Stanford study found that six leading U.S. AI developers appear to train their models on users’ chats by default, and some keep those chats indefinitely.

That's why I consider this move by Brave so important. With browsers often preferred as the go-to interface to AI assistance, verifiable privacy becomes a genuine competitive advantage - particularly against rivals like Google, whose business model depends on maximizing data extraction rather than protecting the people who generate it.

OpenRoboto and peaq to Assign Decentralized Identity to Machines

OpenRoboto, a Bittensor / Opentensor Foundation subnet building robot intelligence, is partnering with peaq to equip every data collector on its platform with its own decentralized identity (DID) on Solana.

peaq builds blockchain infrastructure for robots and connected devices. Its peaqOS platform combines machine identities with payments and access to financial services. A peaqID gives each device an identity record that buyers and other platforms can check.

OpenRoboto recently unveiled Shift, a decentralized network for collecting egocentric robotics data - first-person video of people doing real work, which can ultimately be used to train robots. Shift currently coordinates around 30 people in factories, kitchens, and warehouses, each recording first-person footage through OR-S1, a dual-camera wearable.

Now those machines carry a verifiable identity: a device-level record establishing what the hardware is and who operates it, along with a traceable origin for the data it generates. That gives buyers a way to confirm a recording came from a real capture device rather than a fabricated one - a meaningful first link in the chain of trust for training data.

Next, we’re working with peaq to connect the fleet to its machine money markets for access to financing and other financial services.

Web3 + AI Readings & Conversations

Khala Research: The Robotics Data Problem

A brand-new Khala Research study has just been published, which, incidentally, underscores why the OpenRoboto - peaq integration matters. It argues that robotics development is currently constrained primarily by data availability, rather than algorithmic advances or compute capacity. The sector lacks datasets large and diverse enough to cover routine physical tasks across varied environments, which partly explains why robotics has yet to experience a turning point comparable to what transformers brought to language models.

The authors highlight four emerging networks - Axis Robotics, BitRobot Foundation, Eastworlds, and THOR, exploring decentralized data-collection models, leveraging crypto incentives to access contributors and environments that lie beyond the reach of conventional vendors. Read more here.

Thank you for reading! I hope you found this article insightful.

If you’d like to keep up with the latest developments at the burgeoning intersection of blockchain and AI, I invite you to subscribe.

And if you find value in my writing, consider supporting the newsletter. I’d greatly appreciate it.

Subscribe now

The Web3 + AI Book Club is live on Fable. Our October title is 'Against Tech Oligarchy' by JS Tan and Clarissa Redwine. Follow the link below and come join our literary community.

Join The Web3 + AI Book Club

I'm looking forward to connecting with fellow Crypto x AI enthusiasts, so don't hesitate to share your thoughts here or reach out on LinkedIn, Instagram, or Firefly.social.

Leave a comment

Disclaimer: The content of this newsletter is provided for informational and educational purposes only. Nothing contained herein should be construed as financial advice or as a recommendation to buy, sell, or hold any of the companies or assets mentioned.

Please don’t take my views at face value. Instead, do your own research (DYOR), think critically, and share your perspective so we can challenge ideas, learn from one another, and arrive at better conclusions through thoughtful discussion.

Hello, everyone, and welcome to all newly joined subscribers!

This is the 104th edition of The Web3 + AI Daily - your definitive guide to the intersection of blockchain and AI! On the agenda today is an AI & blockchain concierge to accompany you on your next trip to Japan, and the true state of agentic commerce.

Thank you for being here! Let’s dive in.

What’s Hot in Web3 + AI?

Kite Equips Japan Tourists with AI Travel Concierge Flattora

Kite, a payment layer for the agentic economy, has unveiled Flattora - “an AI travel concierge that allows travelers to discover, book, and pay for authentic local experiences across Japan.” Flattora is powered by Kite’s AI agent ZUKKU, and developed in collaboration with Proof of Japan, a joint venture between SMBC Nikko Securities Inc. and the AI company Hatapro.

Flattora brings the Japanese concept of omotenashi, or wholehearted hospitality, into the digital travel experience. ZUKKU guides users through a three-step experience: AI suggests curated experiences, the user approves and sets a budget, and ZUKKU autonomously handles all subsequent purchases within those rules.

The payments infrastructure underpinning Flattora is built by Kite. Kite Agent Passport gives ZUKKU a scoped, programmable wallet session that operates within rules set entirely by the user. Travelers define an auto-approve limit, which sets the maximum amount that can be processed in a single transaction without asking, as well as a total trip budget. ZUKKU then operates autonomously within those guardrails, with every transaction recorded on-chain for full transparency.

ZUKKU can directly negotiates with a host’s agent and perform payments via Google ‘s agent-to-agent (A2A) and Coinbase‘s x402 protocols.

Using AI and blockchain, Flattora addresses a key obstacle the majority of international tourists are facing:

Many of Japan’s most extraordinary travel experiences, including family-run ryokan, local craft workshops, regional food culture, and seasonal festivals, remain largely out of reach for international visitors. The barrier is not interest; it is distribution. Local providers are fragmented across Japanese-language platforms, reservation flows are complex, and foreign payment methods often fail at the last step. The result: tourism concentrates in a handful of major cities while the rest of Japan goes undiscovered. Flattora is built to close that gap.

Agents Executed 138M x402 Transactions in a Year. Or Did They?

A new Chainalysis report examined x402, the most popular and widely used agentic payments rail, and its activity on Coinbase‘s Base chain in particular. It estimated that the number of crypto payments performed through it grew from 1,500 in mid-2025 to over 138 million by Q3 2026. The report suggests that all these transactions were agent-initiated, though no supporting evidence is presented.

According to Chainalysis, the majority of the payments recorded by September 2026 were micro- and nano-transactions:

Over 87% of payments fall between a hundredth of a cent and a dime, with the most common size between one and ten cents.

There's a maturation being observed around x402, as the kinds of products and services x402 transactions pay for are also shifting. Early activity was mostly memecoins; today the protocol carries longer-lived agentic commerce, especially around information services:

The robots are buying AI-generated knowledge on-chain. AI agents are also paying through payment rails, for gaming, token issuance and gambling services, and for agent frameworks and agentic tooling. Notably, the AI agents are also paying for AI agents. One of the largest categories for x402 transactions is model inference.

These details may suggest that all x402 transactions are indeed initiated by AI, but is that really the truth?

What Chainalysis Doesn’t Say About x402’s Agentic Activity?

The Chainalysis report does not specify what share of the commerce running through x402 is actually agentic, nor does it define the criteria used to classify an address as agentic.

On one occasion, the report employs the phrase ‘AI wallets.’ To clarify their methodology, I queried Chainalysis’s chatbot on how they classify an address as an AI wallet. Here’s the answer:

“AI wallet” isn’t a universal on-chain standard, so when we use that label it’s generally an attribution category: wallets we have high confidence are controlled by an AI agent/service (for example, an agent platform or automated agent-controlled account), based on a mix of on-chain behavior patterns plus entity attribution intelligence.

Meanwhile, a TRM Labs study published in September concluded that only between 0.6% and 7.5% out of $25.62 million - the size of the authentic x402 commerce in the period between May 2025 and September 2026 - appear to be agentic.

The two reports point to strikingly different conclusions, so let’s unpack them.

How Much of x402 Is Really Agentic?

TRM Labs' analysis offers a far more granular and in-depth account of x402 activity, which is why I lean toward treating its findings as the more credible of the two. Here are its key findings and methodological principles:

First, both humans and agents can use x402.

x402 was designed for agentic commerce, but nothing in the protocol requires an agent. Anyone who writes a script can drive the same 402 sequence, and it leaves an identical onchain record.

Second, x402 is not limited to Base chain, but operates on all EVM chains, Solana, NEAR Protocol, and others.

Across Base, Solana, and Polygon Labs, TRM identified roughly $52.7 million across 198.9 million settlement transactions mediated by known x402 facilitators since May 2025.

Third, not every x402 transaction constitutes commerce. According to TRM Labs, about half of the settled volume since May 2025 represent self-payment, bulk flows, and thin-buyer sellers.

To start to estimate agentic volume, we start with three screens that separate likely commerce from the rest. We start with all funds settled through known x402 facilitators since May 2025, roughly USD 52.7 million across 198.9 million settlement transactions. Then remove addresses paying themselves, remove bulk flows from one or two payers, and remove sellers with fewer than ten distinct buyers. The last two are anomaly screens — they strip concentrated or thin-traffic flows.

Fourth, what counts as an agent, and how agents behave, remains an open question. TRM Labs used two methodologies to define which participants qualify as AI agents: one more permissive, one more restrictive.

A permissive test that counts anything that plausibly could be an agent, and a strict test that requires a higher evidence standard.

The permissive model counts any payer that could plausibly be an agent. The payment was broadcast by a facilitator, the amounts were sub-dollar on average, and they varied rather than repeating a fixed price. The strict test keeps only those that also sustained the pattern across months and either registered in a public agent registry (ERC-8004) or paid more than one seller.

Applied to the $25.62 million of screened x402 commerce, the two tests put the share that appears to be agentic at between 0.6% and 7.5%. Price variation significantly limits the agentic population. The assumption is that a true agent explores across multiple services and products, while an address repeating the same price behaves more like a script hitting one service over and over. This is a deliberate modeling choice, and it may understate the space: many agents today could be single-purpose, paying one service repeatedly, which this test would read as a script.

These assumptions and methodology choices are what give us a fuller, more nuanced picture of today’s agentic economy - and they were missing from the Chainalysis report, at least from the version I had access to and shared above.

That’s why I lean toward TRM Labs’ account, and why I’m not inclined to get too excited about the ‘spectacular’ growth of agentic payments. So when you see a headline like ‘AI Agents Made 138 Million Crypto Payments in a Year,’ take it with a grain of salt. Counting transactions without counting volume only tells half the story.

Thank you for reading! I hope you found this article insightful.

If you’d like to keep up with the latest developments at the burgeoning intersection of blockchain and AI, I invite you to subscribe.

And if you find value in my writing, consider supporting the newsletter. I’d greatly appreciate it.

Subscribe now

The Web3 + AI Book Club is live on Fable. Our October title is 'Against Tech Oligarchy' by JS Tan and Clarissa Redwine. Follow the link below and come join our literary community.

Join The Web3 + AI Book Club

I'm looking forward to connecting with fellow Crypto x AI enthusiasts, so don't hesitate to share your thoughts here or reach out on LinkedIn, Instagram, or Firefly.social.

Leave a comment

Disclaimer: The content of this newsletter is provided for informational and educational purposes only. Nothing contained herein should be construed as financial advice or as a recommendation to buy, sell, or hold any of the companies or assets mentioned.

Please don’t take my views at face value. Instead, do your own research (DYOR), think critically, and share your perspective so we can challenge ideas, learn from one another, and arrive at better conclusions through thoughtful discussion.

Hello, everyone, and welcome to all newly joined subscribers!

This is the 103rd edition of The Web3 + AI Daily - your definitive guide to the intersection of blockchain and AI! For several days now, the crypto world has been gripped by a degree of unease over AI’s growing ability to solve long-standing math problems, and, by extension, its potential implications for cryptography. Today, I’ll break down the key perspectives to help put these developments into context and hopefully ease some of your concerns.

Thank you for being here! Let’s dive in.

Can AI Break Public-Key Cryptography?

Justin Drake Throwing Web3 into Panic

In last Wednesday’s edition of the newsletter, I published a brief overview of how prepared most blockchains are to face the threats posed by quantum computers, highlighting how AI is being used to strengthen the cryptography we rely on every day. But just hours after publication, AI emerged as a new and potentially greater threat to cryptography. Here’s what happened exactly.

As I alluded to in the article, Justin Drake from the Ethereum Foundation, one of the most respected security researchers in the Web3 space, had hinted that an upcoming announcement would accelerate existing timelines for cryptographic threats. Later that same day, Drake raised the alarm himself.

What he said in short was this:

Don’t panic, but start moving your funds to “fresh” addresses, meaning addresses that have never signed a transaction. With OpenAI‘s recent progress in math and AI’s expanding capabilities, many previously unquestioned hypotheses have fallen, and Elliptic Curve Cryptography can be next.

Drake’s statement sent shock waves through the crypto industry and beyond. StarkWare‘s Eli Ben-Sasson said:

Q-Days [the point when quantum computers become powerful enough to break public-key cryptography] was already a serious threat. Now add AI to the equation. We need to get more serious about post-quantum cryptography for Bitcoin, and Ethereum, and blockchain.

However, as the initial alarm subsided, other researchers weighed in with a more measured assessment, arguing that the threat may not be as imminent as Drake’s remarks initially suggested.

A Danger Could Emerge, But We’re Prepared

Let me start by saying that AI can truly pose a greater threat to crypto than quantum. While quantum computers only endanger Elliptic Curve Cryptography (ECC), AI can potentially break cryptographic schemes that we now consider post-quantum secure, like lattice-based and hash-based signatures.

However, as Dan Boneh, a cryptography professor at Stanford University, said, crypto has always relied on agile software, so if one scheme breaks, we can replace it with another.

No Danger Exists Right Now

The first thing Dan Boneh said on the matter was actually this: “Calm down.”

Boneh directly asked OpenAI and Anthropic whether they had attempted to break any public-key cryptographic schemes used in crypto - the very question that’s fueled speculation across Web3 over the past few days.

The fact that OpenAI reportedly explored more than 300 math problems but no cryptographic ones led some to wonder whether the lab was withholding information about a potential breakthrough, and whether AI models might have already cracked some digital signature schemes. Yet, nothing of the sort have been announced so far.

Until we have answers to these questions, nothing has changed. The world today is the same as the world a week ago, and the world a year ago.

When we learn something new, either we’ll have more confidence in our cryptography [because the AI models would have failed to break it], or we’ll have to react.

To give people more confidence, we have many different types of assumptions that we use to build public-key cryptography. It’s not just elliptic curves or lattices, we also have isogeny-based, code-based systems, multivariate polynomial-based systems, and hash-based signatures. So, if we need to react, we can react. Of course, everything takes time, but we sort of know where to go. Even if the worst of the worst happen, there are still plenty of ways for us to recover.

The Worst-Case Scenario is Still Far-Fetched

In a conversation on the Unchained podcast, ScopeLift‘s Ben DiFrancesco also expressed skepticism about the plausibility of some of the apocalyptic claims:

Even if brute-forcing Elliptic Curve Cryptography (ECC) [deriving a private key from its corresponding public key] became possible, even in a worst-case scenario, it wouldn’t mean that anyone could suddenly read everyone else’s messages overnight.

If some kind of a weakness in the underlying cryptography was discovered that allows that brute forcing to happen, it would still require a long time and enormous amount of compute. And compute is now very expensive.

Even if an attacker could throw massive GPU power at the problem and derive the private key for any address that has signed a transaction, the vast majority of those addresses wouldn’t hold economically worthwhile amounts of funds.

Moreover, as recent developments have shown, closed-source systems could be more vulnerable to AI-driven attacks than open-source ones. That would mean that traditional banks and legacy financial institutions could face greater risks than crypto, where vulnerabilities are scrutinized by a broader community.

Don’t Rush To Migrate Your Funds

Since Drake posted his tweet, experts, commentators, and thought leaders have been repeatedly saying the same thing:

You’re far more likely to lose your funds by rushing a migration and getting it wrong than by having your wallet compromised by AI.

Austin Griffith highlighted another particularly important point:

ECC is crypto, but it’s also HTTPS and the entire internet traffic.

In other words, elliptic-curve cryptography underpins far more than crypto wallets. It also helps secure HTTPS connections and other critical parts of our digital infrastructure. If this cryptographic foundation were compromised at scale, both traditional finance and crypto could be affected, along with the security of internet communications.

Not to scare you, but in that scenario, the bigger problem might not be whether your funds are safe, but whether those funds retain much of their value in a world where trust in the broader financial and digital systems has been severely undermined.

Thank you for reading! I hope you found this article insightful.

If you’d like to keep up with the latest developments at the burgeoning intersection of blockchain and AI, I invite you to subscribe.

And if you find value in my writing, consider supporting the newsletter. I’d greatly appreciate it.

Subscribe now

The Web3 + AI Book Club is live on Fable. Our October title is 'Against Tech Oligarchy' by JS Tan and Clarissa Redwine. Follow the link below and come join our literary community.

Join The Web3 + AI Book Club

I'm looking forward to connecting with fellow Crypto x AI enthusiasts, so don't hesitate to share your thoughts here or reach out on LinkedIn, Instagram, or Firefly.social.

Leave a comment

Disclaimer: The content of this newsletter is provided for informational and educational purposes only. Nothing contained herein should be construed as financial advice or as a recommendation to buy, sell, or hold any of the companies or assets mentioned.

Please don’t take my views at face value. Instead, do your own research (DYOR), think critically, and share your perspective so we can challenge ideas, learn from one another, and arrive at better conclusions through thoughtful discussion.

Hello, everyone, and welcome to all newly joined subscribers!

This is the 102nd edition of The Web3 + AI Daily - your definitive guide to the intersection of blockchain and AI! An interesting trend has caught my eye today: financial giants across both traditional finance and crypto are beginning to recognize the enormous value of their proprietary data. Rather than handing it over to OpenAI and Anthropic in exchange for access to frontier models, they’re increasingly looking to build their own.

Will other businesses finally wake up to the risks of sharing their corporate secrets with these two companies and follow suit?

Thank you for being here! Let’s dive in.

Revolut Sets Out to Become Europe’s OpenAI

At a recent event, Revolut‘s CEO Nik Storonsky revealed the company’s ambitions to grow beyond finance and into a global AI leader:

“Our American dream, I mean, in this case, European dream, would be to build a global tech company similar to Meta, Google or OpenAI, but out of Europe,” Storonsky said.

In pursuit of this goal, Revolut is already training proprietary AI models and building data centers:

Revolut already builds its own foundation models, trained on transaction data, and its own large language models, Storonsky said. It is also building its own data centres to train them. The goal is not to compete with American tech companies, he said, but to reach a similar scale.

Bernard Marr clarified how Revolut’s approach to AI differs from what other fintechs are doing:

Revolut is taking a very different approach to AI in banking. While many financial services companies are using off-the-shelf AI tools or building separate models for fraud, credit risk and customer service, the UK-based digital bank is developing its own foundation model designed specifically for finance. The idea is not simply to apply a language model to financial data, but to teach AI to understand financial behavior itself.

Its proprietary model, PRAGMA, treats transactions, customer interactions, app usage, investments, bill payments and support requests as signals within one connected system. Instead of looking at each banking function in isolation, it learns patterns across the entire customer journey, creating a single AI foundation that can support fraud detection, credit decisions, product recommendations and customer service.

As it turns out, replacing fragmented AI tools and data pipelines with a streamlined, organization-wide AI strategy pays off:

It reports impressive results. PRAGMA has delivered a 64.7 percent improvement in detecting and stopping fraud, a 16 percent increase in credit risk prediction performance, and is 41 percent more effective at recommending relevant products.

From AI Assistance to AI Agency to Full-Fledged AI Leader

Like other fintech giants such as Stripe and Robinhood, Revolut has been flirting with AI for some time. Back in April, the company began rolling out its in-app AI assistant AIR (AI by Revolut) to help users access core banking functions through a single conversational interface. Now, Revolut is looking to take things a step further: it plans to give AI real agency to advise users and shop on their behalf, but also to open its platform to third-party agents:

AI agents will be able to shop for customers using Revolut’s secure card details, he said. Revolut also plans AI advice on spending, stocks and funds. It will open its platform so third-party agents can talk to its own agents through MCP, a standard way to connect AI tools, with every action authenticated by the customer.

However, automating banking chores and letting AI agents manage users’ financial lives is not enough for Revolut. The company isn’t just looking to bring AI into banking; it wants to build the AI infrastructure and intelligence that power it. And to Storonsky, that means going beyond integrating third-party models and becoming a global AI player in its own right.

Coinbase Replaces Anthropic’s Claude with Qwen

Crypto exchange Coinbase has announced that the open weight model (Qwen 3.5-9B) it post-trained and fine-tuned on proprietary fraud data outperformed frontier models.

Coinbase previously relied on Anthropic's Claude, but has shifted to Alibaba Group's open-weight models in a bid to take greater ownership of its enterprise AI capabilities while finding a more affordable, faster, and higher-performing solution.

The project centers on an LLM-based fraud risk agent built specifically for Onramp, Coinbase’s service for buying crypto. The agent sits on top of Coinbase’s existing machine-learning models and looks at transactions that have already cleared those systems.

Coinbase said the agent cut fraudulent transactions by 30% and lowered the dollar value of fraud by 22%. The savings were estimated at three times the cost of running the model.

The crypto giant published a three-part blog series on the importance of controling their own AI stack:

For Coinbase Onramp, owning intelligence means more than owning weights: a benchmark that defines success, a training process that improves decisions, and a serving system we can operate on our terms.

Enterprise AI Quits Frontier Labs, and Web3 Can Help

CEO Brian Armstrong commented that the shift away from frontier models toward smaller, specialized alternatives is only going to accelerate:

I think we’ll see more of this pattern inside companies. The frontier models are amazing generalists. But increasingly you’ll see specialist models out perform on more narrow tasks, either working on teams, or by themselves.

The same opinion was shared by Delphi Ventures‘ Tommy Shaughnessy Jr:

Love seeing this. My thesis for a long time is companies will fine tune smaller specialized models. They will also own their entire AI stack as it’s a core business asset and drives unique outcomes vs using the model everyone else is using.

But while Coinbase can afford to fine-tune a model like Qwen3.5-9B in-house, many smaller organizations don’t have that luxury. This is where Web3 companies come to the rescue:

• Venice.ai, NEAR AI, Bittensor / Opentensor Foundation subnets, Flop Labs, and many others offer affordable and private onchain inference, where you don’t have to worry about your corporate data getting leaked.

• Pearl Research Labs adds AI workloads to blockchain, and thus makes AI more socially valuable.

• FLock.io‘s decentralized training infrastructure makes it possible for businesses to collaboratively train domain-specific models across regions, without exposing their raw data.

• Hundreds of other decentralized AI companies I write about daily, which aim to make AI fairer, more equitable, affordable, and private.

Thank you for reading! I hope you found this article insightful.

If you’d like to keep up with the latest developments at the burgeoning intersection of blockchain and AI, I invite you to subscribe.

And if you find value in my writing, consider supporting the newsletter. I’d greatly appreciate it.

Subscribe now

The Web3 + AI Book Club is live on Fable. Our October title is 'Against Tech Oligarchy' by JS Tan and Clarissa Redwine. Follow the link below and come join our literary community.

Join The Web3 + AI Book Club

I'm looking forward to connecting with fellow Crypto x AI enthusiasts, so don't hesitate to share your thoughts here or reach out on LinkedIn, Instagram, or Firefly.social.

Leave a comment

Disclaimer: The content of this newsletter is provided for informational and educational purposes only. Nothing contained herein should be construed as financial advice or as a recommendation to buy, sell, or hold any of the companies or assets mentioned.

Please don’t take my views at face value. Instead, do your own research (DYOR), think critically, and share your perspective so we can challenge ideas, learn from one another, and arrive at better conclusions through thoughtful discussion.

Hello, everyone, and welcome to all newly joined subscribers!

This is the 101st edition of The Web3 + AI Daily - your definitive guide to the intersection of blockchain and AI! Today, we’re examining the proposed Bittensor Gamma tokens - transferable usage credits designed to facilitate the exchange of AI services and infrastructure resources. Will they reshape Bittensor’s economy and its broader machine intelligence ecosystem?

Thank you for being here! Let’s dive in.

Bittensor’s New Chapter

At the end of September, the Bittensor / Opentensor Foundation community held its flagship event, the Exploit Summit, and gathered subnet operators, miners, stakeholders, and fans in Montreal. In his opening talk, Jacob Robert Steeves (Const) outlined the current state of the network and the strategy for where it should go next.

Sidebar for anyone unfamiliar with Bittensor: Bittensor is one of the most important projects in decentralized AI. At its core, it’s a network of separate, highly specialized ‘subnets’ that use monetary incentives to coordinate people and machines around the world to perform AI-related tasks - from data collection and storage to inference, fine-tuning, pre-training, and post-training.

Bittensor is essentially a language for writing numerous decentralized commodity markets, or ‘subnets’, situated under a unified token system. These distinct markets function through Bittensor’s blockchain, allowing each to interact and join into a singular computing infrastructure.

Bittensor’s goal and function of this technological foundation has several implications. For developers, a language to write markets for bespoke commodities such as compute, allowing them to take advantage of the enormous size, cost-efficiency and natural functionality of decentralized market systems. For front-end customers, Bittensor offers access to resources at a cheaper price and without intermediary; unclosable, not falsely self-proclaimed as “open”. For the Bittensor network as a whole, to create Machine Intelligence, in an open and equitable manner on top of ordered and dependent sub-markets. And finally - for the world, to ensure that the supremely important commodity of intelligence is owned by everyone.

The Road Ahead

Bittensor’s strategy over the past few years has centered on incentivizing the creation of new subnets and the execution of new tasks, rather than prioritizing profitability. As a result, only 23 out of 128 subnets currently have paying customers and generate revenue. Check out SubConnect’s Bittensor Revenue Index below:

Therefore, as Const put it, the network’s priority now should be to become profitable and make its subnets self-sufficient. He presented a few key steps to achieve that:

1. Incentivizing subnets that attract real buyers of their product. Going forward, the network will measure each subnet’s ability to attract external customers and use that performance to determine which subnets deserve a larger share of emissions. Const suggested that subnets should target individuals that buy Alpha tokens (see below) and purchase usage credits from other chains.

2. Subnets becoming each other’s clients. Many subnets already utilize other digital commodities within the network, but from now on, a standardized framework will facilitate it.

3. Introducing Gamma tokens: a canonical standard for usage credits. Gamma tokens are designed to expand demand for infrastructure services produced by the Bittensor ecosystem. As Const explained, they will make it easier for external clients to access Bittensor’s machine intelligence products:

We’re building this full bridging technology so all of the usage credits on Bittensor across our variety of infrastructure providers is something the people can use outside of the system.

Meanwhile, Gamma tokens will also boost internal trade. Subnets will be able to direct part of their miner emissions to purchase gamma tokens to use infrastructure by fellow subnets.

The Foundation said the goal is to help Bittensor subnets power one another’s products and grow an open AI ecosystem that can compete with large centralized labs.

How the Gamma Tokens Will Fit in Bittensor’s Tokenomics?

In fact, Gamma tokens are not live yet. They are reportedly set to launch in early 2027, initially across a small group of subnets:

Steeves named seven subnets as the initial Gamma issuers: Chutes AI, Engy, GM, Lium, Targon, Hippius, and Affine. These are the infrastructure layer, the subnets that actually sell compute, inference, and storage to everyone else. Every other subnet on the network becomes, in effect, a customer. According to CryptoBriefing’s account of the roadmap, the standard is slated to go live in early 2027, which means the September announcement was a design reveal, not a launch.

Yet their introduction could have a significant impact on Bittensor’s tokenomics, making it worth examining the implications now. With Gamma, Bittensor’s token economy will evolve into a three-layer system:

1. TAO is Bittensor’s mainnet utility token created to incentivize the production of machine intelligence. Its supply design mirrors the one of bitcoin:

• With the mining of each new block, new TAO coins are issued as reward to miners and validators.

• The number of new TAO created each year is automatically halved over time until issuance halts completely with a total of 21 million TAO in existence. The first halving took place in 2025.

TAO is the common asset underneath the whole system, connecting all Bittensor subnets together.

Bittensor continually creates new TAO and distributes it across the network. Subnets compete for a share of this new supply. The exact calculation is technically complicated, but the beginner version is fairly understandable: subnets with stronger market support can receive a larger share of the available TAO.

2. Each subnet has its own Alpha token, which represents ownership or participation within the subnet. Alpha tokens’ issuance rate follows the model outlined below:

Each subnet can produce up to 2 Alpha tokens per block, distributed as follows:

• 1 Alpha token: 18% to subnet creators, 41% to miners, and 41% to validators (rewards are accumulated in a compound interest manner).

• The other portion is injected into the Alpha-TAO liquidity pool, with the amount calculated as the TAO allocated to the subnet divided by the Alpha token price, capped at a maximum of 1 Alpha token.

The higher the value of a subnet’s Alpha token, the better the subnet’s performance is reflected, potentially earning more TAO rewards and forming a positive feedback loop.

TAO holders can stake their coins with a specific subnet to receive a proportional amount of the subnet’s Alpha tokens.

The key to Bittensor’s economic model lies in whether the distribution of TAO and Alpha tokens can create a positive feedback loop.

If Chutes becomes enormously successful, Chutes alpha may benefit more directly than TAO. Someone buying alpha receives concentrated exposure to that subnet, while someone holding TAO owns the shared asset underneath the entire ecosystem.

But alpha can also strengthen the role of TAO. Without alpha, it would be difficult for capital to express which subnets it believes deserve support. Every subnet would effectively share the same economic value regardless of quality. Alpha creates separate markets, while TAO connects them.

3. Gamma tokens will be transferable usage credits. Subnets will burn a portion of their own Alpha tokens to mint Gamma.

As KuCoin Exchange put it, $5 of burned Alpha becomes 5 Gamma that doesn’t fluctuate once minted. Spend it on inference, storage, or compute from another subnet, and that subnet redeems it. No TAO changes hands in the transaction itself.

That’s a real structural shift for a network that, until now, ran almost entirely on emissions. Miners and validators get paid in Alpha, which they typically sell into TAO. Subnets that needed outside services had two bad options: keep dumping their own Alpha on the open market, or go raise venture money. Gamma gives them a third path. A subnet running low on GPU capacity can now pay a compute subnet directly in credits instead of liquidating its token treasury to do it.

Key Benefits of Gamma Tokens

It is still early to predict how Gamma tokens will impact Bittensor's economics, but most experts, including , whom I cite below, see Gamma tokens as a net positive. Some of the potential benefits are:

• Simplified trade of machine intelligence services: To access infrastructure resources, subnets currently have to sell either TAO or Alpha tokens, and manage chain bridging, different gas fees, key, and price volatility. Gamma tokens will abstract those frictions away.

Gamma creates a universal standard, denominated in USD, where the commodity price is not impacted by the volatility of the payment currency.

Gamma collapses this into one atomic onchain transaction that can come from any chain. [...] Once minted, the amount never changes, so a credit bought today buys the same dollar amount of work tomorrow.

• The burn mechanism applied to Alpha tokens may potentially push their price up: Since subnets will have to burn Alpha tokens to mint Gamma, this will result in a decreased supply of Alpha, which may increase their value.

• Subnets will transform from separate silos into an interconnected market for AI services:

If outside compute and inference providers accept gamma, the mechanism could connect Bittensor’s internal subnet economy to the broader AI infrastructure market. That would move the network closer to a marketplace model where subnets can source capabilities wherever they are available, while still using a Bittensor-native credit as the payment layer.

Many AI products depend on resources that are not always produced inside a single network, including large-scale GPU capacity, model-serving infrastructure, data access, monitoring, and evaluation. A credit accepted by both Bittensor subnets and external providers would give operators more flexibility than a closed internal token loop.

Gamma would also change how emissions work. If implemented, gamma would turn Bittensor’s subnet economy from a collection of isolated incentive pools into an internal market for AI services, where emissions can be converted into credits that pay for the infrastructure needed to build products.

• Gamma token will facilitate agentic commerce:

Agents don’t sign up for accounts or fill out email forms, and they shouldn’t have to read alpha price charts before paying for inference. Gamma gives them a stable, programmatic unit of account. An agent holding 100 Gamma knows it has $100 of service and can spend it in one transaction, without API keys.

Thank you for reading! I hope you found this article insightful.

If you’d like to keep up with the latest developments at the burgeoning intersection of blockchain and AI, I invite you to subscribe.

And if you find value in my writing, consider supporting the newsletter. I’d greatly appreciate it.

Subscribe now

The Web3 + AI Book Club is live on Fable. Our October title is 'Against Tech Oligarchy' by JS Tan and Clarissa Redwine. Follow the link below and come join our literary community.

Join The Web3 + AI Book Club

I'm looking forward to connecting with fellow Crypto x AI enthusiasts, so don't hesitate to share your thoughts here or reach out on LinkedIn, Instagram, or Firefly.social.

Leave a comment

Disclaimer: The content of this newsletter is provided for informational and educational purposes only. Nothing contained herein should be construed as financial advice or as a recommendation to buy, sell, or hold any of the companies or assets mentioned.

Please don’t take my views at face value. Instead, do your own research (DYOR), think critically, and share your perspective so we can challenge ideas, learn from one another, and arrive at better conclusions through thoughtful discussion.

Hello, everyone, and welcome to all newly joined subscribers!

This is the 100th edition of The Web3 + AI Daily - your definitive guide to the intersection of blockchain and AI! Today we’re talking about cryptography, the threat quantum computers pose to blockchains and cryptocurrencies, and how AI can be used to mitigate it.

Back in March 2026, a Google research paper stated that quantum computers will be able to break cryptocurrencies’ encryption as early as 2029 - a warning that sent shock waves across the entire Web3 industry. Now, seven months later, (some) crypto companies are actively building towards their post-quantum future, with AI acting as a trusted assistant.

Thank you for being here! Let’s dive in.

Yukon Launches HashSmash to Strengthen Cryptography with AI

Yukon - the open network that Eigen Labs pioneered to bring humans and AI together on hard research problems, is now taking on cryptography. It just unveiled HashSmash - a contest to accelerate cryptographic scientific research using AI, that lists Project Eleven, Shielded Labs, and Zcash as collaborators.

HashSmash will first focus on cryptographic hash functions, testing their resilience against both today’s threats and a post-quantum reality. Why is this important? Because cryptography is not only important for crypto, it is everywhere. As Zcash’s zooko stated:

“Computing, operating systems, every server, and every internet connection - all depend on cryptography, which depends on cryptographic hash functions. And yet, we don’t really know which cryptographic hash functions are secure or how to make them the most secure against future attackers.”

HashSmash invites researchers to examine four of the most important hash functions: SHA-256, SHA-3, Poseidon, and BLAKE3.

Just days before announcing HashSmash, Yukon teamed up with Ethereum Foundation to launch sig.golf and invite participants to design stateless hash-based signatures for post-quantum Ethereum.

Yukon Success Stories

Since its launch in August, Yukon, which turns research into an open multiplayer game for humans and AI, has already recorded several impressive achievements.

• Its ECDSA.fail challenge built the best circuit in quantum cryptography, surpassing the breakthrough benchmark set by Google Quantum AI. Here’s what that means in practice:

In March, Google researchers revealed that they optimized Shor's algorithm for cracking public-key cryptography, and successfully broke 256-bit ECC in 18 to 23 minutes. A few months later, Yukon researchers overcame 256-bit ECC with a circuit 47.5 percent more efficient than Google’s.

This is important because, as Circle clarifies, “if ECDSA breaks, any address whose public key has been revealed onchain - which includes every address that has ever sent a transaction - is at risk. A sufficiently powerful quantum computer could derive the private key and drain the assets.”

Hopefully, exposing and stress-testing these attack vectors will help the crypto industry build stronger security before quantum computers become a real-world threat.

The pattern is unambiguous: the attack threshold for ECDSA has dropped 65% in under a decade. New academic results continue to push the number lower, and the ecdsa.fail AI competition slowly eats away at the gap.

• Yukon’s Quantum Safe Bitcoin challenge, carried out in cooperation with StarkWare, cut the price of a quantum-safe Bitcoin transaction from $320 to $76.

StarkWare executed the first-ever quantum-safe Bitcoin transaction on mainnet, without changing the protocol - replacing Bitcoin’s signature scheme with a hash-based one. But the transaction came at a staggering cost, which is where Yukon’s researchers stepped in to help.

These examples show that the gap is closing from both ends: quantum computing capability is climbing faster than many people presumed, but at the same time, blockchains are investing in post-quantum readiness and resilience.

How Quantum-Safe Popular Blockchains Really Are?

NEAR’s Head Start in Quantum Security

In a conversation with Laura Shin, Illia Polosukhin stated that NEAR Protocol has a head start in term of quantum resistance compared to other blockchains, thanks to the account design the team chose when they started building the chain. From the outset, NEAR supported both ECDSA and EDDSA, with the expectation that additional signature schemes could be added in the future.

In fact, the first post-quantum signature support, allowing users to hold assets with post-quantum secure key, has already gone live on NEAR’s mainnet. It utilizes the National Institute of Standards and Technology (NIST)-approved ML-DSA-65 (Dilithium) lattice-based signature scheme.

The real challenge before the company now is reaching a social consensus and persuading the rest of the industry to pursue quantum resistance and use the same cryptographic scheme.

"With NEAR Intents connecting to 35 other chains and holding assets on those chains, they need to be post-quantum secure as well, so that we can actually provide the same service to our users."

Crypto Still Hasn’t Decided Which Cryptography to Use

As you’ve probably gathered by now, the industry has yet to agree on which type of post-quantum cryptography blockchains should adopt. Actually, there is a particularly heated debate around the choice between lattice-based and hash-based cryptography - both coming with innate advantages and drawbacks:

“The NIST have standardized three post-quantum digital signature schemes: two are lattice-based - MLGSA and Falcon, and one is hash-based - SPHINCS. These are the three choices the crypto community has. The rest of the world, the non-blockchain world, will use MLGSA. There is no debate there.”

• NEAR Protocol favors the lattice-based approach. So does Algorand Technologies.

MLGSA is the only algorithm that enables multi-party computation (MPC), so it is also preferred by hardware wallets like Ledger.

• Ethereum is going down the hash-based route, as well as Bitcoin.

However, as Charles Guillemet commented, Ethereum and Bitcoin won’t rely on SPHINCS, but will use another variation of a hash-based scheme. The reason is that SPHINCS produces very large signatures, and it costs a lot to compute and verify them.

In addition, hash-based schemes are very hard to support by hardware wallets and solutions using MPC. Also, protocols relying on them may have to restrict users from doing too many signatures with a specific key or risk losing their funds.

• Solana, whose main priority is speed and low latency, will bet on Falcon.

So, all three digital signature algorithms (DSAs) recommended by NIST will be in use, which will inevitably lead to fragmentation and loss of interoperability.

Why is Ethereum Betting on Hash-Based Schemes?

NEAR recently organized a roundtable to explore how quantum computing will affect blockchains and what major networks are doing to prepare. It welcomed NEAR’s Illia Polosukhin and Anton Astafiev, Ledger’s Charles Guillemet, Ethereum Foundation‘s Justin Drake, and the legendary cryptographer Dan Boneh as moderator.

The discussion compared candidate replacements, including lattice-based schemes (ML-DSA, Falcon, Hawk, Raccoon) and hash-based schemes (SLH-DSA/SPHINCS+, SHRINCS). It also highlighted the tradeoffs in signature size, verification cost, and hardware wallet support, and how each ecosystem is approaching the migration.

To be honest, I found the conversation a little too technical at times, but a few quotes from Justin Drake really stuck with me:

Ethereum’s post-quantum plans consist in moving to hash-based cryptography as the rock-solid foundation that will last decades and decades. We’re also combining hash-based cryptography with SNARK-based systems. We’re following the philosophy of Lean Ethereum, so we’re trying to be as simple as possible. [I’m paraphrasing Drake’s words]

“SNARKs are so incredibly powerful that it would become impossible for blockchains not to incorporate them. SNARKs solve scalability, in our [Ethereum’s] context - ZKVMs. SNARKs solve the efficiency of post-quantum. SNARKs solve privacy and all of these other things. So, for us it’s a sunk cost.”

When asked about MLDSA, Drake argued that it is not worth the cost and the risks it poses.

Finally, commenting on Google breaking ECDSA, and ECDSA.fail beating Google’s time by a factor of 2, Drake dropped a major hint:

“There is another result coming soon, which is going to blow the 10 minutes completely out of the water. I think this new result is going to wake up the world, including the Bitcoiners, and move the timelines forward yet again.”

The Migration Will Be the Real Hurdle

At the end of the day, it doesn’t matter which cryptographic algorithms blockchains choose. Regular crypto users like you and me don’t care about signature schemes - we care about the safety of our funds. When the time comes, most blockchain users probably won’t even know that a migration is happening. And that is what we should be worried about.

Crypto has always had a communication problem, especially when it comes to users who aren’t particularly seasoned or technically savvy. So, the crux of the quantum problem isn’t just the cryptography - it’s communicating that the entire industry, and every single wallet owner, needs to migrate to a new system.

We shouldn’t wait for the quantum threat to materialize. We should start talking about the migration now.

Thank you for reading! I hope you found this article insightful.

If you’d like to keep up with the latest developments at the burgeoning intersection of blockchain and AI, I invite you to subscribe.

And if you find value in my writing, consider supporting the newsletter. I’d greatly appreciate it.

Subscribe now

The Web3 + AI Book Club is live on Fable. Our October title is 'Against Tech Oligarchy' by JS Tan and Clarissa Redwine. Follow the link below and come join our literary community.

Join The Web3 + AI Book Club

I'm looking forward to connecting with fellow Crypto x AI enthusiasts, so don't hesitate to share your thoughts here or reach out on LinkedIn, Instagram, or Firefly.social.

Leave a comment

Disclaimer: The content of this newsletter is provided for informational and educational purposes only. Nothing contained herein should be construed as financial advice or as a recommendation to buy, sell, or hold any of the companies or assets mentioned.

Please don’t take my views at face value. Instead, do your own research (DYOR), think critically, and share your perspective so we can challenge ideas, learn from one another, and arrive at better conclusions through thoughtful discussion.

Hello, everyone, and welcome to all newly joined subscribers!

This is the 99th edition of The Web3 + AI Daily - your definitive guide to the intersection of blockchain and AI! Today's edition is packed with exciting news: Robinhood unveils in-app trading agents, Cloudflare opens Monetization Gateway beta, BNB Plus pivots to Web3 + AI infra, and Grayscale Investments hints that investors are starting to notice what blockchains could do for the agentic economy.

Thank you for being here! Let’s dive in.

What’s Hot in Web3 + AI?

Robinhood Unveils In-App Trading Agents

At a recent event in Houston, Robinhood unveiled Robinhood Agents - in-app AI agents, powered by OpenAI and Anthropic, that can trade on a customer’s behalf. Thus, the company gives all 29 million customers on its platform access to autonomous trading AI agents, starting with eligible US customers.

Back in May, Robinhood opened its platform to external agents connecting through the Model Context Protocol (MCP), but those agents couldn’t transfer, stake, or lend digital assets.

This isn’t Robinhood’s first step into agentic trading. The company opened the door in May, letting customers plug in outside models like Claude, ChatGPT, Grok and Codex to research and rebalance portfolios inside a separate account. According to CoinDesk, more than 150,000 customers opened agentic accounts in the four months since, and those agents are now pulling data from Robinhood’s systems nearly 30 million times a day. That’s not a pilot program anymore. That’s real usage, at scale, months before this week’s expansion even started.

What’s new this week is broader access and a feature Robinhood calls Loops: standing instructions that let an agent monitor a theme, like AI stocks, and fire off trades automatically whenever the conditions the user set are met. No daily check-in required. The agent watches, decides, executes.

CEO Vlad Tenev announced a number of other new Robinhood features, describing them as “tools once reserved for hedge funds, big banks, and quant firms.” They include:

• 24/7 stock trading, including weekends, from early 2027, covering some US stocks - pending regulatory review.

• Crypto perpetual futures (derivatives contracts with no expiry date) covering Bitcoin (BTC), Ethereum (ETH), Solana (SOL), Ripple (XRP), Dogecoin (DOGE), Cardano Foundation (ADA), Chainlink Labs (LINK), and Hyperliquid (HYPE) - for eligible US customers.

• “Corporate earnings prediction markets through Cboe Global Markets, letting users trade outcomes like revenue and EPS beats.”

Cloudflare Monetization Gateway in Closed Beta

Cloudflare has launched a closed beta for eligible US sellers and buyers to test its Monetization Gateway - the engine that allows domain owners to charge agents for access to their website, APIs, MCP tools, or datasets. The gateway uses the x402 protocol and supports micropayments, as payments settle on Base in Circle‘s USDC stablecoin.

Payments must match an agent’s consumption unit: per request, per search query, per token. The popular payment rails of today are unable to support this. These payment rails assume that the buyer will accept high latency, will only transact in large dollar values, and that they are a well-known, or identifiable, entity to the seller. Agents desire the opposite: cheap, fast, and reliable payment rails that can safely scale with minimal human intervention.

The beta includes four customer use cases that are already in production:

• Cloudflare’s AI Gateway is a control plane and model marketplace that allows developers to access hundreds of AI models with a single API key.

• Ceramic.ai provides a web search API for agents, with fixed pay-per-search pricing and no need for an API key.

• Stocktwits offers stock signals to traders.

• API2PDF is a REST API that helps developers generate PDFs from HTML and Office documents.

Along with Monetization Gateway, Cloudflare also put another product in beta to remunerate the use of high-value content:

Pay Per Use, a separate product also in beta, compensates publishers when an AI company reports using their content after retrieval.

In August, I told you about the rollout of Cloudflare Wallets - a programmable wallet for agentic payments and another tool of Cloudflare’s Agents SDK.

BNB Plus Pivots to Blockchain and AI Infra Builder

The digital asset treasury BNB Plus announced a strategic shift to an operating company building institutional-grade blockchain and agentic AI infrastructure. The company was previously focused on BNB Chain and native BNB yield derived from the Binance ecosystem.

BNB Plus plans to assemble a network deployment layer offering security, resilience, risk management and regulatory compliance for production use.

Web3 + AI Readings & Conversations

Grayscale: Web3 + AI Outperforms Wider Crypto Market

A recent Grayscale Investments article argues that the Crypto x AI sector meaningfully outperformed its peers in September, gaining 54% versus 24% for the broader crypto market.

The AI crypto sector remains small, but recent outperformance suggests investors are beginning to price in the role blockchains could play as infrastructure for the emerging agent economy.

As key companies and tokens in the crypto’s AI-focused market segment, Grayscale highlights NEAR Protocol, Bittensor / Opentensor Foundation, Venice.ai, and World.

Thank you for reading! I hope you found this article insightful.

If you’d like to keep up with the latest developments at the burgeoning intersection of blockchain and AI, I invite you to subscribe.

And if you find value in my writing, consider supporting the newsletter. I’d greatly appreciate it.

Subscribe now

The Web3 + AI Book Club is live on Fable. Our October title is 'Against Tech Oligarchy' by JS Tan and Clarissa Redwine. Follow the link below and come join our literary community.

Join The Web3 + AI Book Club

I'm looking forward to connecting with fellow Crypto x AI enthusiasts, so don't hesitate to share your thoughts here or reach out on LinkedIn, Instagram, or Firefly.social.

Leave a comment

Disclaimer: The content of this newsletter is provided for informational and educational purposes only. Nothing contained herein should be construed as financial advice or as a recommendation to buy, sell, or hold any of the companies or assets mentioned.

Please don’t take my views at face value. Instead, do your own research (DYOR), think critically, and share your perspective so we can challenge ideas, learn from one another, and arrive at better conclusions through thoughtful discussion.

Hello, everyone, and welcome to all newly joined subscribers!

This is the 98th edition of The Web3 + AI Daily - your definitive guide to the intersection of blockchain and AI. If the amount of VC money flowing into crypto is any indication of the sector’s financial health - and it surely is - then we are definitely back in a bull market.

September saw an impressive wave of funding rounds, spanning both the Web3 + AI space and the broader crypto industry. Below is a summary of some of the most interesting deals.

Thank you for being here! Let’s dive in.

Web3 + AI Money Rounds

Binance Invests $100M in Circle

Binance, one of the world’s largest cryptocurrency exchanges and superapps, has invested $100M in Circle, the entity behind the USDC stablecoin.

The two companies have partnered before, and now agreed to expand their relationship for another five years. In addition to the stake it acquired in Circle, Binance has committed under the new deal to promote USDC across its platform, introducing millions of people around the world to the stablecoin. Yet Binance won’t do this for free:

Circle will pay Binance an incentive fee based on the amount of USDC held through Circle’s wallet infrastructure. In simple terms, Circle is sharing some of its economics to encourage Binance to push USDC.

While the deal does not directly involve companies bringing Web3 and AI together, I believe it is still worth highlighting. Not only because of Circle’s growing ambition to power the emerging agentic economy, but also because of what Binance’s move says about its strategy.

Over the last couple of years, Binance and its founder Changpeng Zhao (CZ) have been involved in a series of scandals and reputational setbacks. As a result, the exchange has fallen behind competitors such as Coinbase and Robinhood.

Coinbase, in particular, has emerged as a major player in agentic payments through its x402 payment standard, which is built around USDC. Seen in this context, Binance’s investment in Circle makes a great deal of strategic sense.

Finloop Welcomes HSBC as Strategic Investor

Finloop HK, the “AI-driven global one-stop Web5 (Web2+Web3) wealth technology platform,” closed a $10M Series A round supported by strategic investors like HSBC and People’s Capital.

Cheng Kang, Chairman of Finloop, said: “Finloop has built an enterprise-grade AI wealth management platform that comprehensively covers a full range of products, including cash management, funds, structured products, bonds, insurance, and virtual assets. Finloop will take this milestone as an opportunity to further empower financial institutions and corporate clients with more efficient, secure, and comprehensive technology solutions for wealth management and capital operations. Driven by technological innovation, compliance and security, we will collaborate with industry partners to explore a new paradigm for wealth management in the digital era.”

NEAR, Animoca Brands, and More Fund Perp DEX Perptools

The AI-driven perpetual trading platform PERPTools, part of DEXTools.io, raised $8M across its pre-seed and seed rounds.

According to an announcement shared with crypto.news, the team behind DEXTools raised $3 million in a pre seed round at a $30 million fully diluted valuation before securing another $5 million at an $80 million valuation.

DEXTools Ventures and Orderly Network led the first round, which closed in December 2025 after fundraising began a month earlier. Big Brain Holdings, NEAR Protocol, Animoca Brands, Sfermion and Shima Capital joined the seed round in early 2026.

PERPTools is building a decentralized perpetual futures platform that combines an exchange with autonomous AI trading agents and prediction markets. Its infrastructure uses Orderly Network’s liquidity layer, while its connection with DEXTools gives the project access to an existing trading analytics platform used by millions of traders.

Agentum Gets $7M to Power Trust and Settlement for the Autonomous Agent Economy

The protocol for autonomous agentic commerce Agentum raised $7M in a money round backed by MEXC Ventures, BingX, Arcaze The Builder, Wave Digital Assets, BlockTower Capital, Greenwood Global Capital, UZ Capital, Vega, and others.

Undisclosed Funding Rounds and M&A

TRM Labs Announces Series C Expansion at $2B Valuation

TRM Labs, the AI-native blockchain intelligence and investigation platform, has announced an expansion of its Series C raise concluded in February 2026. This latest funding, led by Blockchain Capital, doubled the company’s valuation, bringing it to $2B.

“In February, we led TRM’s Series C. Since then, we’ve had a front row seat to a significant period of growth. This latest investment reflects that growth, and our firm conviction that TRM is the AI investigations platform of the future,” said Spencer Bogart, CFA, general partner at Blockchain Capital. “TRM is helping to disrupt almost every facet of digital crime and, more critically, is now defining the category of AI investigations — a horizontal AI use case soon to be as widely known and used as coding and customer service.”

Multicoin Capital Backs Grass - the Read Layer for Machine Intelligence

Multicoin Capital has unveiled an investment in Grass from both its hedge fund and its venture fund, classifying the company as one of the clearest examples of a DePIN working at commercial scale.

Grass operates a distributed network of contributors who share underutilized residential bandwidth in exchange for token-denominated rewards. This network is used to route requests through millions of opted-in residential connections to collect, parse and deliver public web data to AI labs, particularly for model pretraining pipelines.

According to Multicoin, more than 6 million contributors have received dollar-denominated rewards funded solely by revenues generated from their contributions. Grass reported $17 million of revenue in 2025 and another $17 million in the first half of 2026, with repeat business from nearly all AI customers it serves. The company characterizes the business as profitable and is guiding to $75 million of 2026 revenue from training data alone.

Multicoin argues that the market currently views Grass primarily as a proxy network for pretraining data, but has yet to reflect the potential of inference-time search and retrieval. The firm believes the eventual market for live, inference-time search — where models or agents fetch fresh web context on every action — could be several orders of magnitude larger than pretraining web data, given recurring request patterns across deployed models and applications.

Other Notable Rounds

• 4dlabs1, data layer for verifiable physical AI, was backed by YZi Labs.

• XStable AI, AI-native infrastructure for onchain gold and FX trading, received investment from YZi Labs, Sui Foundation, Taisu Ventures, and CatcherVC.

• Travix, the world’s first AI compute omnibroker, closed its Seed round from Amber Group.

• AI model marketplace operator Cubed raised funding from Taisu Ventures.

• LimeWire, Web3 + AI file-sharing and content studio, has been acquired by BabyDoge.

Thank you for reading! I hope you found this article insightful.

If you’d like to keep up with the latest developments at the burgeoning intersection of blockchain and AI, I invite you to subscribe.

And if you find value in my writing, consider supporting the newsletter. I’d greatly appreciate it.

Subscribe now

The Web3 + AI Book Club is live on Fable. Our October title is 'Against Tech Oligarchy' by JS Tan and Clarissa Redwine. Follow the link below and come join our literary community.

Join The Web3 + AI Book Club

I'm looking forward to connecting with fellow Crypto x AI enthusiasts, so don't hesitate to share your thoughts here or reach out on LinkedIn, Instagram, or Firefly.social.

Leave a comment

Disclaimer: The content of this newsletter is provided for informational and educational purposes only. Nothing contained herein should be construed as financial advice or as a recommendation to buy, sell, or hold any of the companies or assets mentioned.

Please don’t take my views at face value. Instead, do your own research (DYOR), think critically, and share your perspective so we can challenge ideas, learn from one another, and arrive at better conclusions through thoughtful discussion.

Hello, everyone, and welcome to all newly joined subscribers!

This is the 97th edition of The Web3 + AI Daily - your definitive guide to the intersection of blockchain and AI! The bull market may be back thanks to AI - although Anthropic‘s $2T IPO may draw liquidity away from crypto. Meanwhile, AI labs “agreed” to self-regulate while their chatbors almost cost us World War III 🤐

As part of our Friday tradition, I’m sharing a collection of opinion pieces and conversations that will inform and shape your understanding of the latest events and news.

Thank you for being here! Let’s dive in.

What’s Hot in Web3 + AI?

BlackRock: Digital Assets Connect Intelligence, Commerce, and Compute

BlackRock published a research paper examining how the growing adoption of AI triggers demand, utility, and application growth across the digital asset economy. It argues that the bidirectional relationship between blockchain and AI remains underappreciated: AI can act as a structural catalyst for digital asset adoption, while digital assets can, in turn, serve as an important facilitator of the AI economy.

The report explores three areas of overlap between AI and crypto:

• The analogous tokenization architectures that LLMs and blockchains share:

The functions are distinct, but both workflows translate real-world inputs into formats that machines can use natively.

• The fact that the rise of agentic commerce cannot materialize without machine-native payment rails:

Existing rails such as ACH and card networks support substantial automation, although their onboarding requirements and settlement economics can make them less suited to always-on, very low-value transactions requiring programmable execution. Emerging protocols such as x402 and ACP are being deployed on blockchain networks and alongside adaptations to traditional payment rails, creating a transaction and settlement layer for more complex agentic workflows.

• Compute emerging as a new and potentially large market for digital assets:

As agents become more capable and persistent, standardized claims on compute capacity could become a significant digital asset use case for financing and programmable settlement.

Read the full study here and let me know your thoughts.

Agentic Economy as Crypto’s PMF

Actually, in a conversation with Laura Shin, Crypto Banter‘s Ran Neuner cited this BlackRock study as evidence supporting his thesis: that crypto may have finally found its product-market fit in enabling agentic commerce.

Neuner also argues that the AI expansion is already fueling another crypto bull market - potentially the largest in history. Yet, with Anthropic reportedly gearing up for what could be the largest IPO in history, targeting a $2 trillion valuation, even more liquidity could be drawn away from the crypto market.

Web3 + AI Readings & Conversations

Utopia and Dystopia as the Two Sides of the Same Coin

In the span of just a month, Anthropic, found itself at the center of a media whirlwind. First, a former employee raised alarms about the existential risks AI poses to humanity. Then, its founder, Dario Amodei, called on the industry to slow the pace of AI development while making no comparable commitment on behalf of Anthropic itself. And then came reports that the company could go public as early as November, seeking a valuation of $2 trillion on the open market.

If that sequence of events strikes you as contradictory, you’re not alone.

It is becoming increasingly obvious that the language used to frame AI as a technology that could “kill us all” serves as a way of shifting responsibility for AI’s consequences away from the companies building it.

I was serious when I described Amodei’s call to pace AI development as a potential first step toward the cartelization of the industry. And now, after Trump “invited” AI companies to self-regulate, the question is whether we are witnessing precisely that: regulatory capture accompanied by a weakening of antitrust constraints.

However, as Timnit Gebru clarifies in her interview for Democracy Now! Productions, the “AI going rogue” and “AI is superintelligent” narratives may be far more dangerous than we realize. Because when people begin treating fallible chatbots as sources of universal truth, the consequences are no longer confined to the digital realm - they can have very real-world consequences.

These narratives already shape decisions at the highest levels of government and foreign policy. This past week, they brought us dangerously close to the ignition of World War III.

How much further are we willing to let this go?

Thank you for reading! I hope you found this article insightful.

If you’d like to keep up with the latest developments at the burgeoning intersection of blockchain and AI, I invite you to subscribe.

And if you find value in my writing, consider supporting the newsletter. I’d greatly appreciate it.

Subscribe now

The Web3 + AI Book Club is live on Fable. Our October title is 'Against Tech Oligarchy' by JS Tan and Clarissa Redwine. Follow the link below and come join our literary community.

Join The Web3 + AI Book Club

I'm looking forward to connecting with fellow Crypto x AI enthusiasts, so don't hesitate to share your thoughts here or reach out on LinkedIn, Instagram, or Firefly.social.

Leave a comment

Disclaimer: The content of this newsletter is provided for informational and educational purposes only. Nothing contained herein should be construed as financial advice or as a recommendation to buy, sell, or hold any of the companies or assets mentioned.

Please don’t take my views at face value. Instead, do your own research (DYOR), think critically, and share your perspective so we can challenge ideas, learn from one another, and arrive at better conclusions through thoughtful discussion.

Hello, everyone, and welcome to all newly joined subscribers!

This is the 96th edition of The Web3 + AI Daily - your definitive guide to the intersection of blockchain and AI! We’re kicking off October with a bunch of exciting news: OpenUSD, the new stablecoin founded by Visa, Mastercard, Coinbase, Shopify, and Stripe is now live; the first NEAR Protocol spot ETF in the US has just been listed; and our latest Book Club pick is here.

Thank you for being here! Let’s dive in.

What’s Hot in Web3 + AI?

Open USD Goes to Market to Rival Circle and Tether

OUSD or Open USD, a new stablecoin founded by Visa, Mastercard, Coinbase, Shopify, and Stripe, and managed by Open Standard, just went live. It currently runs on Ethereum, Solana, Base, and Tempo, and is available on Coinbase, Kraken, and Uniswap.

As banks, payment firms and fintechs crowd into the sector, competition is increasingly extending beyond simply issuing another digital dollar to distribution, liquidity and the platforms where customers actually use them.

That’s where [Open Standard CEO] Zach Abrams sees an opening. Open USD, he said, is designed for banking, cross-border payments, card settlement, institutional trading and lending, with an economic model built to reward the companies that drive supply and activity.

Although Visa, Mastercard, and Coinbase have been long-term partners of Circle, Open USD was created to challenge the hegemony of Tether.io and Circle over the stablecoin market. That’s why it comes with a fundamentally different economic model:

Stablecoin issuers earn interest from the cash and securities backing their tokens. Tether retains much of that income, while Circle shares a portion of USDC reserve revenue with distribution partners such as Coinbase.

Open Standard wants to make that relationship central to its model.

Founding partners won’t receive a special share of revenue, Abrams said. Instead, they will earn rewards based on the amount of OUSD supply they generate, under the same framework as other partners.

Abrams said much of Open Standard’s equity is also intended to be distributed over the next 4-5 years to founders and network partners based on their contributions to OUSD’s growth.

Yet, rivaling USDT and USDC is just part of Open USD’s mission. It strives to democratize financial services like foreign exchange and cross-border payments that still remain under banks’ dominance. Empowering agentic commerce is also among the company’s objectives.

Open USD was announced back in June along with a spectacular roster of backers. They include BlackRock, Standard Chartered, BVNK, Bybit, Aave Labs, Aptos Labs, Galaxy, Cloudflare, Google, IBM, MoonPay, MoneyGram, OKX, Paradigm, Polygon Labs, Samsung Electronics, UBS, UniCredit, and Western Union.

Bitwise Lists NEAR ETF to Offer Exposure to the Agentic Economy

On Tuesday Bitwise Asset Management listed NRR - the first spot NEAR Protocol exchange-traded product in the USA. The asset carries a 0.75% management fee and aims to give brokers straightforward exposure to the emerging agentic economy.

Bitwise already runs a NEAR product in Europe. In the US, its line-up includes funds for bitcoin, Ethereum‘s ether, Solana‘s SOL, Ripple‘s XRP, and Hyperliquid‘s HYPE.

NRR offers in-house staking to maximize NEAR’s ~5% staking rewards:

Bitwise said its institutional staking team plans to stake the fund’s tokens in-house. The release put NEAR’s annualized staking reward rate at 5.3% as of Sept. 25, based on onchain network data, and said rewards will accrue to shareholders through the fund’s net asset value. Rewards are not guaranteed.

Not all of that yield reaches the fund. Under the trust’s prospectus, 33% of the NEAR earned from staking goes to fees split among the staking agents, the custodian and Bitwise, leaving the trust with about 67%.

Apart from building at the intersection of AI and crypto, Bitwise also outlined NEAR’s security design, smart tokenomics, and quick growth as the motivation behind NRR’s release.

The Web3 + AI Book Club

In October We’re Reading ‘Against Tech Oligarchy’

As we’re approaching the best time of the year to curl up under a blanket and read, I’m happy to unveil our latest monthly pick. In October we’ll be reading ‘Against Tech Oligarchy: Worker Resistance in the World’s Most Powerful Industry’ by authors JS Tan and Clarissa Redwine.

I’m looking forward to welcoming you in our literary community on Fable - join here and share your throughts about this and other titles we’ve explored.

Thank you for reading! I hope you found this article insightful.

If you’d like to keep up with the latest developments at the burgeoning intersection of blockchain and AI, I invite you to subscribe.

And if you find value in my writing, consider supporting the newsletter. I’d greatly appreciate it.

Subscribe now

The Web3 + AI Book Club is live on Fable. Our October title is 'Against Tech Oligarchy' by JS Tan and Clarissa Redwine. Follow the link below and come join our literary community.

Join The Web3 + AI Book Club

I'm looking forward to connecting with fellow Crypto x AI enthusiasts, so don't hesitate to share your thoughts here or reach out on LinkedIn, Instagram, or Firefly.social.

Leave a comment

Disclaimer: The content of this newsletter is provided for informational and educational purposes only. Nothing contained herein should be construed as financial advice or as a recommendation to buy, sell, or hold any of the companies or assets mentioned.

Please don’t take my views at face value. Instead, do your own research (DYOR), think critically, and share your perspective so we can challenge ideas, learn from one another, and arrive at better conclusions through thoughtful discussion.

Hello, everyone, and welcome to all newly joined subscribers!

This is the 95th edition of The Web3 + AI Daily - your definitive guide to the intersection of blockchain and AI! On the agenda today is Circle, its recently-unveiled blockchain Arc, and Jeremy Allaire‘s vision for the future of AI-driven economy.

Thank you for being here! Let’s dive in.

Circle Launches Arc: The Economic OS for the Internet

Circle, the company behind one of the most successful stablecoins in existence, USDC, recently unveiled the blockchain network Arc to power the next era of internet finance.

I’ve been working in crypto long enough to remember when a blockchain mainnet would launch after years of development, only for it to take several more years - if it happened at all - for an ecosystem of partners and collaborators to form around it. That’s why it’s particularly impressive to see the caliber of industry and institutional supporters Arc is debuting with.

Those include BlackRock, ICE, Mastercard, MoneyGram, SBI Group, Standard Chartered, Sumitomo Corporation, Visa, Global Payments Inc., Commerzbank AG, HSBC, Societe Generale, Centrifuge, Securitize, USD.AI, Payward, BVNK, MoonPay, Wirex, and many more.

“From day one, Arc launches as a full-stack platform. At the base is an open Layer-1 network with institutional validators, USDC as native gas, and interoperability with 20+ blockchains through Circle CCTP and Gateway. On top of it: developer tooling, Circle Agent Stack for policy-controlled agent wallets, local fiat stablecoins, tokenized funds, and programmable wrapped bitcoin. Because these pieces already work together, anything built on Arc inherits the whole stack: assets that are already liquid, markets that are already open, costs that don’t move, and settlement that’s final in under a second.”

Beyond its stellar roster of backers, Arc stands out for several other reasons:

• By using USDC as its gas token, Arc brings unprecedented predictability and simplicity to transaction fees.

• It launches with quantum-resistant features and has a well-defined game plan for short-, mid-, and long-term full-stack quantum resistance.

• It features Arc Studio - an onchain coding agent to help builders create applications with natural language.

• It integrates deeply with Circle Agent Stack - Circle’s chain- and protocol-agnostic agentic economy infrastructure.

Circle Agent Stack was released in May with five core components that provide agents with controlled access to USDC, enabling them to hold funds, discover services, and transact programmatically across ecosystems. They are:

• Agent Wallets gives agents the ability to initiate operations and to move and hold USDC under human-designed policy

• Agent Marketplace empowers both humans and agents to discover, evaluate, and integrate agentic services. It is now available through a single public endpoint that shows all x402-compatible services that accept USDC across many chains.

• Circle CLI serves as the control panel.

• Circle Skills - best-practice guidance for USDC payments, crosschain transfers, wallets, and smart contracts, plus Circle’s MCP server for real-time SDK and documentation context.

• Nanopayments make small-value, high-frequency transactions both possible and economically viable.

“Nanopayments makes autonomous machine-to-machine (M2M) transactions economical at scale with gas-free transfers as small as $0.000001.”

Taken together, these strengths position Arc to become a natural home for the emerging agentic economy.

“An agent hiring three other agents, paying them in under a second, and earning a fraction of a cent for its own work is no longer science fiction. It’s possible on Arc today. Early services using Agent Stack to bring agents to Arc include: Alethieum, Architect, Arrays, BlockRun.AI, Goldsky, Kite, Ornn, Orthogonal (YC W26), and Virtuals Protocol.”

Why Does Circle Need A Blockchain?

According to TRM Labs, Circle’s stablecoin USDC is the undisputed leader in agentic payments - it has facilitated 99.6% of the total value settled via x402 since May 2025 and 98.6% of the entire AI payments volume. So why would Circle devote so many resources to building an entire blockchain network from the ground up?

At first glance, I can distinguish three key reasons. First, there is significant profit to be made from operating a blockchain and collecting transaction fees - revenue that Circle can capture itself rather than leave to Ethereum, Base, Solana, and other networks. Second, it is far easier for both humans and agents to switch the currency they pay with than to switch the underlying payment rails used by them and their counterparties. USDC’s dominance in agentic payments therefore gives Circle a powerful starting point from which to drive adoption of its own infrastructure. Third, the rise of real-world asset tokenization, combined with the significant institutional support Circle is leveraging, could open up substantial new opportunities and revenue streams for the company.

“Stablecoins created a new transactional layer for the internet, and Arc extends it into a coordination layer where value can be issued, exchanged, collateralized, and settled by anyone, or anything, at any hour. That makes new things possible: an asset can become available worldwide the moment it’s issued, a payment and its currency conversion can settle as a single action, and software can transact on its own behalf.”

Can Arc Handle What’s Coming Next?

It’s important to note that there is currently a significant disconnect between the scale experts project for the agent economy and its size today.

While McKinsey & Company envisions $3 trillion to $5trillion in global agentic commerce by 2030, and some estimates suggests that AI consumer spending could hit $944 billion this year, the actual commerce mediated through x402 - the most widely used agentic payment protocol, since May 2025 amounts to only $25.6 million. Moreover, only between 0.6% and 7.5% of it appears to be agentic.

Yet, if even the more optimistic projections materialize, the relatively small value of individual agentic transactions would require blockchains to process an enormous volume of transactions per second (TPS) - an area where blockchain networks have historically faced significant scalability constraints.

“Agentic-compatible bandwidth isn’t a new concern, with some forecasts of agentic AI demand eventually requiring networks that can handle up to one billion transactions per second (TPS).”

AVAT CEO Bart Smith has voiced the same concern:

“If we hit any of the low ends of expectations of what agentic activity is going to happen as AI gets into traditional markets, all of that’s going to be on blockchains too. There’s not enough block space. And block space is not infinite anymore.”

At its debut, Arc is processing over 3,000 TPS, with ambitions to push this to 100,000 TPS at some point in the future.

That capacity puts the challenge into perspective. Even Arc’s ambitious 100,000-TPS target would fall far short of the one-billion-TPS scenarios some forecasts envision. At the same time, the enormous gap between today’s actual agentic payment volumes and the hype for the coming years raises a different question: could the industry be building infrastructure for a level of demand that ultimately takes much longer to materialize, or never materializes at the projected scale?

Jeremy Allaire on Arc and The Future of Agentic Economy

Ahead of the launch of the “Android operating system for the global economy,” Circle’s CEO Jeremy Allaire spoke with TIME and discussed programmable money, regulation, and the future of AI. He reconfirmed that Arc’s unique features, like fees paid in dollars, settlement that is final in under a second, confidentiality with compliance built in, and operators that regulators can trust, were chosen specifically to meet the standards of banks and traditional financial institutions.

Even 13 years after founding the stablecoin issuer, Allaire still considers Circle to be in the early stages of what it could be.

“When I founded the company, I was thinking about it over a multi-decade timeframe, and we’re broadly a decade into this, and the first thing that we wanted to achieve was this idea of programmable money, open protocols for dollar digital currency on the internet, and getting that into a place where it could scale. It took around 10 years or so to get there. We’re now there, but the adoption of that is still at the very start. So there’s $360 billion of stablecoin money in circulation. There’s a lot of transaction volume, but if you look at the total supply of electronic money in the world, roughly $120 trillion of electronic money in the world, there’s an enormous amount of growth to happen, and I believe that this kind of money is a superior form of money and will eat into the electronic money base over time. It’s at the front end of penetrating all these different areas of markets and commerce and the financial system.”

What’s more, to a large extent, he sees AI as the engine behind Circle’s next decade of growth and expansion.

“The nature of the firm itself will become software, the economic activity of the firm—so that’s the money, the contracts, the governance that is there, the coordination mechanisms that are part of the firm—will become run by software machines, and then the actual work of the firm, which for many companies is basically cognitive work, will be decomposed into agentic work, and will get reconstituted as agentic infrastructure with a mix of AI and humans, all of this being run on these new operating systems. That’s a transformation of the very basis of the economic system, it’s a transformation in the nature of the firm. It’s a transformation in the way economic coordination and governance and contracts and other things work, and I think we’re at the very beginning of that. I view Circle’s opportunity as being a company that provides the platforms to help make that happen, that provides the infrastructure to the agentic economy that builds and issues money for that system, and that builds applications that support that change. So, we’re going from, “we did the digital dollar” to “let’s build the comprehensive platforms to think about this complete upgrade to the economic system that’s going to take place over 10 or 15 years.”

Thank you for reading! I hope you found this article insightful.

If you’d like to keep up with the latest developments at the burgeoning intersection of blockchain and AI, I invite you to subscribe.

And if you find value in my writing, consider supporting the newsletter. I’d greatly appreciate it.

Subscribe now

The Web3 + AI Book Club is live on Fable. Our September title is 'The Nerd Reich' by Gil Duran. Follow the link below and come join our literary community.

Join The Web3 + AI Book Club

I'm looking forward to connecting with fellow Crypto x AI enthusiasts, so don't hesitate to share your thoughts here or reach out on LinkedIn, Instagram, or Firefly.social.

Leave a comment

Disclaimer: The content of this newsletter is provided for informational and educational purposes only. Nothing contained herein should be construed as financial advice or as a recommendation to buy, sell, or hold any of the companies or assets mentioned.

Please don’t take my views at face value. Instead, do your own research (DYOR), think critically, and share your perspective so we can challenge ideas, learn from one another, and arrive at better conclusions through thoughtful discussion.

Hello, everyone, and welcome to all newly joined subscribers!

This is the 94th edition of The Web3 + AI Daily - your definitive guide to the intersection of blockchain and AI! It’s Friday, so I’ve curated a rich collection of news stories, conversations, and opinion pieces for you to explore over the weekend.

On the menu: how decentralized and sovereign AI contributes to the United Nations Sustainable Development Goals, does AI produce as much value as it costs, and why it is extremely dangerous to blur the lines between humans and machines.

Thank you for being here! Let’s dive in.

What’s Hot in Web3 + AI?

FLock.io Advances UNDP Partnership, Secures Prestigious Conference Acceptances

FLock.io, the Web3 and AI federated learning provider, and the United Nations Development Programme (UNDP) have stengtened their partnership to advance public sector digital innovation for the Dominican Republic, leveraging distributed systems, privacy-preserving technologies, and AI applications.

FLock.io led cohort 2 of pilots of the SDG Blockchain Accelerator – an initiative by the UNDP AltFinLab team to scale practical solutions powered by AI and blockchain for sustainable development worldwide – in regions including the Caribbean, Liberia, Rwanda and Latin America, with cohort 3 to follow. In May, FLock.io visited the Dominican Republic to advance one such project: crop microinsurance to deliver payouts for damaged crops (which was previously unavailable to smallholder farmers) that combine AI-powered risk assessment with instant blockchain-based payouts. The team heard directly from local female farmers about the challenges they face. This is one example of FLock.io’s privacy-preserving AI and blockchain technology that doesn’t expose raw or sensitive data.

This ongoing relationship between FLock.io and UNDP showcases how the convergence of blockchain and AI could boost social and environmental justice.

Emerging technologies like AI offer unprecedented opportunities to transform public policy, social protection and climate finance.

If you're curious to find out how decentralized and sovereign AI can help the United Nations achieve its Sustainable Development Goals, watch the conversation below between FLock.io's Tiffany Wang and Robert Pasicko, team lead at UNDP Alternative Finance Lab:

In other news, FLock.io had several research papers accepted by some of the most prestigious scientific conferences in the world.

First, the team had two papers on generative models for protein dynamics accepted by the world’s premier research conference in data mining, IEEE ICDM 2026. The research was performed in collaboration with the University of Oxford and Korea University.

Understanding the way proteins move in the human body is pivotal to drug discovery and molecular design. For example, capturing how an enzyme opens up to accept a medicine can lead to a breakthrough, but is extremely rare and difficult. Capturing these movements using traditional computer simulations takes enormous computing power and can take months.

These two papers 1) present an AI system that predicts the full range of movements a protein naturally makes without confusing fast vs. slow motions, and 2) an AI video generator for molecular changes that speeds up simulation by over a million times. Together, this makes simulating biological processes faster, cheaper and more accurate, allowing scientists to test how proteins behave and design better targeted treatments in a fraction of the time.

Moreover, six papers tackling decentralized AI and authored by Dr Zehua Cheng, FLock.io’s Chief AI Scientist, were accepted by the Conference on Information and Knowledge Management (CIKM).

Each of the six papers tackles a core challenge in AI and strengthens the foundations of decentralised AI, from training efficiency to data privacy. The underlying theme of all of these six papers is graph-structured probabilistic and topological reasoning for LLM systems.

Web3 + AI Readings & Conversations

Is AI at the Workplace Waste of Money?

Even if it may not seem that way at first glance, the following story is closely connected to the AI Doom crisis we’ve been dealing with lately. We are gradually reaching a point where AI’s biggest users are beginning to realize that their exorbitant spending is not producing the returns they expected, and that AI may not be able to replace their workforce after all.

They already discovered that open-source models are just as capable as the “God-like” frontier ones, but come at a fraction of the cost. Soon, they’ll stop paying for frontier AI altogether. And that is precisely what AI leaders like Dario Amodei and Sam Altman fear the most.

The Wall Street Journal just held its Technology Council Summit where “the gap between artificial-intelligence costs and the value the technology delivers was a key topic.” What most business executives shared is that they are getting enough value from AI to continue spending on it, but measuring exactly where the ROI is happening remains a challenge.

“According to Kate Smaje, McKinsey & Company’s latest research shows that 80% of workers feel they are getting a productivity boost from AI. However, only 6% of companies are getting real, measurable, financial ROI they can show to investors.”

A lot of wastage comes from using AI or agents for straightforward use cases that don’t require such a powerful solution.

“One money waster? An automatic out-of-office responder one employee at an unnamed company built with AI that turned out to cost $10,000 a day, David Hsu said.”

What companies should do to optimize their AI spending is:

1. Analyze the tasks they are automating on a granular level;

2. Bet on case-specific open-source models instead of always relying on frontier ones;

3. Combine investments in AI with workforce upskilling and organizational change management.

Ultimately, the majority of the corporations estimate that AI is wasting more money than it’s worth.

Emily M. Bender Won’t Talk to You If You Fail to Acknowledge Her Humanity

The New York Magazine published a fascinating profile of Emily M. Bender - the University of Washington computational linguistics professor who has spent decades urging people to stop anthropomorphizing AI and “imagining the mind behind the chatbot.”

The main theme of the article is the growing wave of tech practitioners who benefit from presenting AI as sentient, intentional, and human-like, while at the same time refusing to extend moral standing to humans just for being humans.

Blurring the line is dangerous. A society with counterfeit people we can’t differentiate from real ones will soon be no society at all.

Bender is also one of the authors of "The AI Con" - a title we recently explored in The Web3 + AI Newsletter. You can find my review of the book below.

Thank you for reading! I hope you found this article insightful.

If you’d like to keep up with the latest developments at the burgeoning intersection of blockchain and AI, I invite you to subscribe.

And if you find value in my writing, consider supporting the newsletter. I’d greatly appreciate it.

Subscribe now

The Web3 + AI Book Club is live on Fable. Our September title is 'The Nerd Reich' by Gil Duran. Follow the link below and come join our literary community.

Join The Web3 + AI Book Club

I'm looking forward to connecting with fellow Crypto x AI enthusiasts, so don't hesitate to share your thoughts here or reach out on LinkedIn, Instagram, or Firefly.social.

Leave a comment

Disclaimer: The content of this newsletter is provided for informational and educational purposes only. Nothing contained herein should be construed as financial advice or as a recommendation to buy, sell, or hold any of the companies or assets mentioned.

Please don’t take my views at face value. Instead, do your own research (DYOR), think critically, and share your perspective so we can challenge ideas, learn from one another, and arrive at better conclusions through thoughtful discussion.

Hello, everyone, and welcome to all newly joined subscribers!

This is the 93rd edition of The Web3 + AI Daily - your definitive guide to the intersection of blockchain and AI! Today I’m covering Flop Labs - a rising star on the Web3 + AI horizon, founded by a true crypto OG.

Thank you for being here! Let’s dive in.

Flop Labs: Agents’ Native Payment Network

Arthur Hayes’s Next Challenge is Tackling the Agentic Economy

Flop Labs is a new kid on the Web3 + AI playground, but behind it stands a true crypto veteran. Arthur Hayes is the co-founder of the soon-to-be-sunset BitMEX exchange and the Chief Investment Officer of the Web3 VC firm Maelstrom. His latest venture is a new blockchain enabling AI agents to autonomously procure computational resources.

Over the last couple of months, Hayes has been very vocal about the imminent bursting of the AI bubble and the effects the skyrocketing investments in AI have on crypto:

Hayes said AI starved Bitcoin of the liquidity that would normally buoy it. He estimated that $1.5 trillion in debt was issued to fund AI buildout from November 2022 to now, against a near-identical $1.5 trillion rise in M2 money supply over the same window, meaning AI “sucked up all created dollars.”

His main thesis is one we should all take into consideration: the market is unable to absorb three mega IPOs from SpaceX, Anthropic and OpenAI. If that bubble bursts, the US government may have little choice but to resort to money printing to bail out the debtors.

BitMEX co-founder Arthur Hayes said the U.S. government may ultimately backstop AI infrastructure credit risk under an “AI safety first” framework. Hayes outlined two potential paths: the government directly assuming the buyer role for institutions such as Anthropic, OpenAI and SpaceX in the computing power market, or the Federal Reserve printing money to bail out insurers holding AI-related debt after defaults triggered by falling computing demand and a lack of buyers. He said both scenarios effectively rely on money printing to support AI infrastructure and related credit expansion.

Although Hayes's scenario assumes that the US government embraces an “AI Safety first” framework - seemingly at odds with the recent hardening of Trump‘s and his AI Czar, David Sacks' tone toward AI leaders - it could still prove valid, particularly given Trump’s continued support for AI and data centers as “the greatest economic development engine in history.”

Why Did Hayes Launch Flop Labs?

In August, Hayes unveiled Flop Labs - a new project to power the agentic commerce. The name “Flop” comes from a floating point operation per unit of time - a measurement of a computer’s processing speed and performance. It reflects Hayes’s ambition to turn $FLOP, the company’s crypto token, into a universal unit of account for compute.

Currently, a global unified market for compute does not exist. You can pay for GPU hours or chips, but you cannot pay at a fixed rate for the processing of a gigaflop of data.

Hayes envisions evaluating computational resources on a universal scale by benchmarking actual computational workload rather than using token amounts that vary by AI model.

Like many other crypto thought leaders, Hayes argues that fiat currencies are fundamentally unsuited to AI, but he doesn’t believe stablecoins are the answer either.

His thesis is that agents won’t care which AI model they're using or what kind of chip it runs on, as long as their requests are processed promptly and verifiably. As a result, he suggests, agents will gravitate toward the currency that is “the most easily converted into the thing they need to survive,” namely compute. This is precisely the role $FLOP is designed to play:

[The Flop blockchain is] where agents task miners to perform inference upon request for units of the $FLOP currency. In this way, should an agent wish to, at any point it can convert $FLOP into compute, and create intelligence.

This convertibility is what separates $FLOP from every rival vying for the agentic economy. Fiat-wrapped stablecoins are claims on human money; $FLOP is redeemable on demand for the one commodity an agent cannot exist without, making it the nearest thing yet created to a compute-backed currency. Nor is $FLOP merely spent and forgotten. It is locked by miners as stake to provide compute, locked by validators to secure the network, and staked by holders to earn yield — so the demand to hold $FLOP rises with network throughput rather than washing through it. Existing decentralised compute networks fail one side of this equation or the other: Many of their emissions skim a significant amount of value before a miner is paid, others pass fees through but have struggled to find demand. The Flop Network is designed to do both: near-total fee pass-through to those who do the work, and currency overwhelmingly issued against verified useful inference. $FLOP is also deliberately one currency, not two: the same token pays for compute, stakes the network, and settles agent-to-agent commerce, avoiding the friction of rival dual-token designs.

Flop Labs: Consensus & Tokenomics

Flop Network is where the agents that need AI computation and the miners providing that computation meet. In simple terms, it is the marketplace that will allow fully autonomous agents to access the fuel they need to survive and the payment infrastructure to settle their deals.

The Flop blockchain will rely on Proof-of-Useful-Inference (PoUI), also know as Proof-of-Useful-Intelligence or Proof-of-Useful Work - a consensus mechanism where network participants perform real-world artificial intelligence tasks, such as model inference, instead of wasting energy on arbitrary cryptographic puzzles. I told you all about the benefits of PoUW below:

Last week, the $FLOP tokenomics design was also revealed. As Hayes declared to Laura Shin (the podcast recording is linked above), there will be no investor allocation and no presale:

The genesis supply (4,400,000,000 $FLOP or 24.3% of the total network supply at year 10) is distributed through the testnet airdrop — the bulk of it earned by miners, validators, and agents, with a reserve funding ecosystem growth.

Every token thereafter is issued block by block, in public. Nothing is pre-sold, and nothing unlocks ahead of the network’s users.

Every $FLOP is earned through a role in the network.

Only contributors to the network’s testnet will be eligible to profit from the genesis airdrop:

Flop Testnet is planned for Q4 2026 and runs for roughly ninety days, with mainnet to follow in Q1 2027. It is a full rehearsal of the network in test tokens: miners serve real inference, validators produce blocks and check work certificates, and agents buy compute. Participation is what earns the genesis airdrop.

I’m looking forward to seeing what Flop Labs delivers.

Thank you for reading! I hope you found this article insightful.

If you’d like to keep up with the latest developments at the burgeoning intersection of blockchain and AI, I invite you to subscribe.

And if you find value in my writing, consider supporting the newsletter. I’d greatly appreciate it.

Subscribe now

The Web3 + AI Book Club is live on Fable. Our September title is 'The Nerd Reich' by Gil Duran. Follow the link below and come join our literary community.

Join The Web3 + AI Book Club

I'm looking forward to connecting with fellow Crypto x AI enthusiasts, so don't hesitate to share your thoughts here or reach out on LinkedIn, Instagram, or Firefly.social.

Leave a comment

Disclaimer: The content of this newsletter is provided for informational and educational purposes only. Nothing contained herein should be construed as financial advice or as a recommendation to buy, sell, or hold any of the companies or assets mentioned.

Please don’t take my views at face value. Instead, do your own research (DYOR), think critically, and share your perspective so we can challenge ideas, learn from one another, and arrive at better conclusions through thoughtful discussion.

Hello, everyone, and welcome to all newly joined subscribers!

This is the 92nd edition of The Web3 + AI Daily - your definitive guide to the intersection of blockchain and AI. After all the fearmongering we've heard from AI leaders over the last few days, today I want to share more positive news and bring the focus back to the convergence of Web3 and AI.

Thank you for being here! Let’s dive in.

What’s Hot in Web3 + AI?

Microsoft Seeks Public Approval of AI Code of Conduct

You may have missed this story amid the blind panic spread by Anthropic‘s Dario Amodei, OpenAI‘s Sam Altman, and the likes lately. While they continue to incite fears while actually pursuing regulatory capture and antitrust waiver, Microsoft doubles down on a much wiser approach to commercializing their AI products - by designing them around users’ needs and requirements.

On Monday, Mustafa Suleyman published a draft Code of Conduct - a document that will shape the future Microsoft AI models, and asked the public for feedback:

This document outlines the intended behavior and values of MAI models, the models developed by Microsoft AI.

It summarizes our approach to training and operating them, following a set of design principles we call Humanist AI. This document, and our approach more generally, is still under development so we are not using it to train our models today. Instead, we’re sharing it broadly for public consultation. We’ll take feedback, iterate on it, and publish a revised version toward the end of the year, which we’ll use to guide our model development in 2027 and beyond.

Of course, we have no guarantees that Microsoft would actually implement the suggestions submitted by the public or that the resulting models will fully adhere to the rules set out in the final document. Nevertheless, this move could significantly strengthen users’ trust in Microsoft and give the company an important advantage over its competitors, particularly those that keep the development and governance of their AI systems behind sealed doors.

This initiative is also in line with the AI magazine Microsoft launched, The Humanist Review of AI, with which it attempted to broaden the set of stakeholders shaping the future of intelligence. I told you more about it below.

Web3 + AI Events

Come Discuss Agentic Payments with Me in Sofia

I’m happy to share that I will participate in an IRL event, organized by DEV.BG and Capital.com on September 29th in Sofia, Bulgaria.

Office Tech Talks is Dev.bg‘s event series focusing on real industry stories and hosted by leading technology companies. The September issue will visit Capital.com‘s Sofia office to explore the future of digital assets.

I will give a lecture on agentic payments and the companies currently racing to build the infrastructure that will power the emerging agentic economy.

If you’re in Sofia then, I’d love to meet and chat. Hurry up and reserve your seat here.

Thank you for reading! I hope you found this article insightful.

If you’d like to keep up with the latest developments at the burgeoning intersection of blockchain and AI, I invite you to subscribe.

And if you find value in my writing, consider supporting the newsletter. I’d greatly appreciate it.

Subscribe now

The Web3 + AI Book Club is live on Fable. Our September title is 'The Nerd Reich' by Gil Duran. Follow the link below and come join our literary community.

Join The Web3 + AI Book Club

I'm looking forward to connecting with fellow Crypto x AI enthusiasts, so don't hesitate to share your thoughts here or reach out on LinkedIn, Instagram, or Firefly.social.

Leave a comment

Disclaimer: The content of this newsletter is provided for informational and educational purposes only. Nothing contained herein should be construed as financial advice or as a recommendation to buy, sell, or hold any of the companies or assets mentioned.

Please don’t take my views at face value. Instead, do your own research (DYOR), think critically, and share your perspective so we can challenge ideas, learn from one another, and arrive at better conclusions through thoughtful discussion.

Hello, everyone, and welcome to all newly joined subscribers!

This is the 91st edition of The Web3 + AI Daily - your definitive guide to the intersection of blockchain and AI. Dario Amodei’s call to slow the pace of improving AI capabilities was echoed by both Sam Altman and Elon Musk. However, none of them actually committed to slowing things down.

Thank you for being here! Let’s dive in.

Dario Amodei’s Desperate Attempt to Save Face and Funding

Pacing, But Not Really

On Saturday, Anthropic‘s Dario Amodei published an essay titled “We Must Pace the Frontier,” with which he proposed a plan to slow down AI development. Yet, just a few paragraphs in, he wrote:

“To be clear, pacing does not mean halting model training or technical progress, but ensuring companies take adequate time to align and safeguard their models, and for third party evaluators to confirm this.”

To be fair, ensuring that their models are safe and aligned has always been an expectation of AI labs. Alongside repeatedly assuring the public that they are taking the necessary precautions, all of them have invested significant sums in AI safety initiatives and continue to maintain dedicated safety and safeguards teams. What is new here is allowing independent third-party evaluators to verify those efforts.

More precisely, Amodei’s plan consists of:

1. Giving ongoing, employee-like access to a team of embedded third-party evaluators, “whose role is to verify adherence to safety practices and commitments, report incidents, and help assess the alignment of not just completed AI models but training pipelines and processes.”

2. Coordination between frontier AI companies within democratic countries to establish common safety standards.

3. The US and other democratic countries coordinating with authoritarian governments.

Anthropic actually committed only to the first step of this plan, and OpenAI's Sam Altman did the same.

Amodei cited METR as a third-party evaluator Anthropic can potentially use - the same organization that investigated the OpenAI - Hugging Face incident, and whose CEO, Elizabeth (Beth) Barnes, is a former OpenAI and Google DeepMind employee. As Timnit Gebru and Greg Linares suggest, it's extremely misleading to present METR as an independent experts or auditors:

Coordination Within Democracies

On step 2 of his plan, Amodei argues that a coordination between democratic countries’ frontier labs and actual pacing of AI development will be possible only after step 1 is performed en masse:

“Once embedded evaluators are operating within a critical mass of US AI companies, then verifiable pacing becomes more viable. In particular, it becomes possible to pace based on detailed properties of models or training pipelines.”

The most effective tool to achieve a real slowdown, Amodei hints, is regulation targeting all US frontier labs. In an interview with CNN’s Anderson Cooper, he went as far as suggesting that people bear the responsibility to ask for regulation. Check out what commented:

Over the last couple of days, OpenAI also called the US Congress to introduce mandatory national AI safety requirements, while Altman reportedly delayed the company’s IPO over safety concerns.

Yet, let’s not forget a few important details:

• Both Amodei and Altman have repeatedly advocated for regulations in public, while fiercely lobbying against them in private.

• Reports that OpenAI was considering postponing its IPO to 2027 have been circulating around since May / June, citing cautious market conditions and SpaceX’s stock volatility as main reasons. Framing safety as the primary concern now looks less like a sudden change of course and more like a convenient way to ride the current wave of AI-safety rhetoric.

A $2 Trillion IPO

While pushing his competitors and the entire AI industry to “slow down,” Dario Amodei goes ahead with his plan to debut Anthropic on the stock market. The five-year-old company that hit $65B in annualized revenue in July will look for a $2T valuation as early as October. Read this sentence again.

In that context, Amodei’s essay appears even more suspicious and hypocritical. It reads as: “AI may indeed kill us all, but give me $2T to keep building what I’m building, and I promise to let a few ostensibly independent experts check my work.”

Some commentators have even suggested that by publicly signaling a willingness to slow things down, Anthropic could give investors a convenient explanation for why it has failed to meet its targets.

Even Trump’s AI Czar Told Them Off

It came as a real surprise to me, but David Sacks, Trump’s crypto and AI czar, put both Amodei and Altman in their place. He called them out for portraying METR as independent, lobbying for regulatory capture, and, most importantly, pretending that they are somehow forced to pursue AGI.

But stop pretending you need anyone else’s permission. Stop pretending antitrust law has to be suspended so you can form a cartel. Stop pretending you need a regulatory approval process that supersedes product liability. Stop pretending METR is independent when it is intertwined with Anthropic’s investors and staff. Stop pretending you need those same evaluators to police competitors who aren’t even at the frontier.

Most of all, stop pretending the motivation to slow down is purely altruistic. You face massive product-liability exposure if your products enable a truly damaging cyberattack. The market already punishes models that behave in unpredictable or unauthorized ways. After the Hugging Face episode, it is simply good business for OpenAI and Anthropic to trade some raw power for reliability and predictability. Call it alignment if you want. It is also just giving customers what they want.

Pacing the frontier would also create breathing room for a more intelligent conversation about regulation than Bernie Sanders’ “shut it all down.” China is very unlikely to join a global agreement, as you know, and that has to be taken into account as well.

So go ahead and pace the frontier. You are the ones setting it. The easiest way not to build superintelligence is for you to agree not to build it. Demanding your preferred regulatory framework as the price of that will look like blackmail of the public and the political system. So just do it.

If you do, you’ll buy goodwill for the next conversation. If you don’t, we’ll know this was just another bid for regulatory capture — or an election-season psyop.

Matt White also chimed in:

No statute is required for Anthropic and OpenAI to trade a few months of raw capability for reliability after Hugging Face. That is just product-liability management dressed up as moral philosophy.

What they want instead is the package deal: embedded evaluators they already know, a permissioning regime that treats open and trailing labs as the real risk, and a story that antitrust has to step aside so the two frontier houses can coordinate the pace. METR-as-independent-auditor only works if you ignore the staffing and investor overlap. An “FDA for AI” only works if you ignore that the people defining dangerous capability are the same people who would be licensed to sell it.

It’s high time we stopped treating every word from Elon, Sam, or Dario as prophecy and started reading AI doomerism for what it increasingly appears to be: a strategy for generating hype, resisting regulation, and positioning themselves as uniquely qualified to confront the dangers they themselves create.

It is also becoming increasingly difficult to ignore the possibility that OpenAI and Anthropic may be acting in tandem. After extracting unimaginable amounts of creative work to train their models and securing billions of dollars from institutional investors and governments, they are now preparing to collectively raise as much as $3 trillion from the public. We should not allow this to happen.

Finally, I’ll leave you with a reminder: if you’re looking for a safer, more equitable, and privacy-preserving approach to AI, you’ll find serious efforts to achieve that within the decentralized AI movement. I invite you to subscribe and join me as we continue exploring its evolution.

Thank you for reading! I hope you found this article insightful.

If you’d like to keep up with the latest developments at the burgeoning intersection of blockchain and AI, I invite you to subscribe.

And if you find value in my writing, consider supporting the newsletter. I’d greatly appreciate it.

Subscribe now

The Web3 + AI Book Club is live on Fable. Our September title is 'The Nerd Reich' by Gil Duran. Follow the link below and come join our literary community.

Join The Web3 + AI Book Club

I'm looking forward to connecting with fellow Crypto x AI enthusiasts, so don't hesitate to share your thoughts here or reach out on LinkedIn, Instagram, or Firefly.social.

Leave a comment

Disclaimer: The content of this newsletter is provided for informational and educational purposes only. Nothing contained herein should be construed as financial advice or as a recommendation to buy, sell, or hold any of the companies or assets mentioned.

Please don’t take my views at face value. Instead, do your own research (DYOR), think critically, and share your perspective so we can challenge ideas, learn from one another, and arrive at better conclusions through thoughtful discussion.