Hook: The Letter That Could Not Stop a Race
On a quiet Tuesday in late 2024, 1,178 AI practitioners – including chief scientists from OpenAI, Anthropic, Google DeepMind, and Meta – signed an open letter. Its message was terrifyingly simple: the frontier models are moving too fast. They are approaching a threshold where they can autonomously conduct most AI research. The signatories called for an international slowdown mechanism, a pause button for the most advanced systems. The world applauded their courage. But as someone who spent 2017 trying to build a DAO in Cape Town, I read the letter and felt a familiar ache. It is the same ache I felt when CapeHorizon collapsed under gas fees: you cannot pause a race by asking runners to slow down. You need a referee. You need rules that are coded, enforced, and transparent. You need blockchain.
Vibes > Algorithms – but only when the algorithms are built to protect the vibe. The letter is a cry for help from the brightest minds in AI. They are trapped in a prisoner’s dilemma where no single company can afford to slow down voluntarily. The solution, however, is not a centralised treaty that will take years to negotiate. The solution is a trustless coordination layer that already exists: the same technology that powers the crypto economy. In this essay, I will dissect the letter’s core claims, explain why they point directly to a Web3 architecture, and lay out a concrete path forward. This is not a theory. It is a call to action for every builder who believes that code is law, but people are truth.
Context: The Prisoner’s Dilemma of AI Safety
The letter, first reported by Beating, represents an unprecedented consensus: 1,178 individuals from every major AI lab, plus formal endorsements from OpenAI and Anthropic as companies. The core argument is that advanced AI systems will soon be able to recursively improve themselves, performing the work of thousands of PhDs. At that point, the risk of an uncontrolled intelligence explosion – a paperclip maximiser scenario – becomes existential. The signatories demand an international mechanism to synchronise a slowdown of frontier training runs until adequate safety frameworks are in place.
But here is the brutal truth: no single company can afford to be the first to pause. As the letter itself admits, “individually, no company dares slow down first, because they lose competitive advantage.” This is the classic tragedy of the commons, dressed in Silicon Valley clothing. The AI industry is a race to the bottom of safety. Each lab fears that if they stop, a rival will overtake them and capture the market – or worse, build something they cannot control. Without a credible, binding commitment from all players, a voluntary slowdown is a fantasy.
The letter’s authors are not naïve. They know this. That is why they demand an international regulatory framework, likely modelled on the International Atomic Energy Agency (IAEA). Yet, as the analysis of the letter reveals, the signatories deliberately avoided specifying enforcement mechanisms. They did not propose verification methods, threshold triggers, or dispute resolution. This is where the blockchain narrative begins.
Core: Why Blockchain Is the Only Viable Coordination Layer for AI Safety
I have spent 27 years in the crypto space. I have seen DAOs fail due to gas costs, DeFi protocols crash from composability risks, and NFT projects die after the hype faded. I have also seen the same technology unlock global cooperation without central authority. The AI slowdown problem is, at its core, a coordination problem. And blockchain is the best tool humanity has invented for solving coordination problems without trust.
Embrace the volatility, find the signal. The signal here is that a slowdown mechanism requires three things: verifiability, enforceability, and global participation. Centralised treaties fail on all three. A blockchain-based system can deliver each one.
Verifiability
The first challenge: how do you know that a lab is actually slowing down? You cannot rely on self-reporting, because the incentive to cheat is enormous. The letter mentions that AI systems will soon be able to “autonomously conduct most AI research.” Once that happens, a cheating lab could deploy its AI to secretly advance while publicly claiming to pause. The only way to detect such cheating is through verifiable computation.
I remember the DeFi liquidity trap of 2020. I jumped into three yield farming protocols simultaneously, chasing 100% APYs. I made $15,000, but I also accidentally discovered that composability risks could hide huge leverage. That taught me that transparency is not enough – you need on-chain audit trails. For AI safety, we need a system where every frontier training run is recorded on a public blockchain, along with proofs of compute usage. Zero-knowledge proofs (zk-SNARKs) allow a lab to prove that it only used X amount of compute without revealing the model weights. This is already feasible: projects like Succinct Labs are bringing zk-proofs to any computation. If every lab must submit a zk-proof of their training FLOPs to an on-chain registry, then a slowdown becomes auditable.
Enforceability
Verification alone is useless without enforcement. The letter’s biggest gap is the assumption that nation-states will create and enforce a global slowdown. History shows otherwise: every arms race has ended with a treaty that was either never signed or broken. What if instead, the enforcement mechanism is coded into the infrastructure itself?
Consider a smart contract that holds a bond – say, $1 billion in stablecoins – that every participating lab deposits. The contract defines a global compute budget for frontier training runs per quarter. If any lab submits a zk-proof that exceeds its allowance, the bond is slashed and distributed to other participants. This is a cryptoeconomic security model, just like the slashing in Ethereum’s proof-of-stake. The penalty is automatic, transparent, and irrevocable. No need for a UN vote. No need for trust.
I learned this lesson during the NFT cultural renaissance of 2021. I co-founded AfricanCode, a project that connected Cape Town artists with NFT markets. We raised $80,000 in 48 hours, but then the hype faded because we had no sustainable value proposition. The lesson: community alignment must be backed by incentives, not just enthusiasm. A bond-based slowdown mechanism aligns the self-interest of every lab with the collective safety goal. If you cheat, you lose money. That is stronger than any treaty.
Global Participation
The letter specifically calls for the US to lead, with other nations joining later. This is a fatal flaw. China, the EU, and India will not simply follow a US-centric regime. We saw the same problem with the Paris Agreement – it took years to get near-universal buy-in, and enforcement is weak. A blockchain-based system can be borderless by design. Any lab, anywhere, can join by deploying a smart contract on a public blockchain (Ethereum, Solana, or a dedicated L2). The rules are the same for everyone. The verification is done by the network, not by a government.
I think back to the bear market pivot in 2022. My portfolio crashed 70%, but I found refuge in ZK-rollup research. I spent six months studying Succinct Labs, and I realised that the same technology could solve global coordination problems. ZK-rollups scale Ethereum by verifying batches of transactions off-chain. Why not scale AI safety by verifying compute usage off-chain? The infrastructure is already there. We just need to build the application layer.
Contrarian: The Blind Spots of Crypto-Based Safety
Now let me address the elephant in the room: blockchain is too slow, too public, and too rigid for AI safety.
Build in public, live in truth – but sometimes truth is dangerous. If every lab must publish zk-proofs of their compute, an adversary could infer the pace of someone’s research. Even with zk-proofs, metadata leaks. A lab might not want to reveal that it is working on a new training paradigm until it is ready. There is a trade-off between transparency and secrecy. However, the letter itself demands transparency for safety. If a lab wants to keep its research secret, it should not participate in the slowdown regime – and then it becomes a rogue actor. The system can handle that with a whitelist: only labs that commit to the bond are trusted. Others are considered high-risk and may face sanctions from the regulated market (e.g., cloud providers refusing GPU sales).
Another blind spot: latency. AI training runs are highly dynamic. A global bond-based system might introduce delays in submitting proofs, slowing down the pace even more. But that is exactly what the letter wants – a deliberate slowdown. The question is whether the overhead is acceptable. I believe it is, because the alternative is an uncontrolled intelligence explosion. A 10% latency tax is better than extinction.
Third, the assumption that nation-states will accept a private, crypto-led system is optimistic. A government like China might never allow its labs to use an Ethereum smart contract. True. But the system does not need to be universal to be effective. Even a subset of labs – say, the top ten in the US and Europe – can create a safety club. The club can then pressure others through market access. If you want to sell your API to Fortune 500 companies, you must be a club member. This is the same mechanism that led to the adoption of financial audits: not regulation, but market demand for trust.
What the Analysis Missed
The original analysis of the letter gave a B- confidence rating to the technology roadmap. It correctly noted that “autonomous research” is years away, not months. That means we have time. But the analysis also pointed out that the letter hides a governance design blank. My proposal fills that blank with blockchain. I am not claiming that a single smart contract will stop the singularity. I am saying that the letter’s call for an international slowdown mechanism is incomplete without a verification and enforcement layer. Blockchain provides that layer, and the crypto community should start building it now.
Code is law, but people are truth. The code must reflect our shared values. A zk-based compute registry is not a panacea. It needs governance: who decides the compute threshold? How often is it updated? The answer is a DAO of AI labs, academics, and civil society. I know DAOs can fail – CapeHorizon did. But since 2017, the technology has matured. We have L2s that handle thousands of transactions per second. We have on-chain voting with quadratic funding. The tools exist. What is missing is the will to apply them to AI safety.
Takeaway: The Future-Back Call to Build
I have been writing about Web3 since 2017. I have seen the hype cycles come and go. But this moment feels different. The AI safety letter is a desperate plea from the very people building the technology. They are asking for a mechanism to control what they have created. The crypto community has spent years building the infrastructure for trustless cooperation. Now is the time to bridge the two worlds.
I propose a concrete next step: a working group of Web3 developers and AI safety researchers to design a proof-of-concept for an on-chain compute registry. I am willing to fund the first $50,000 from my personal savings, using the same bonds model I described. If you are a developer who knows zk-proofs, or a researcher who understands scaling laws, reach out. Let’s build this before the machines learn to build themselves.
Embrace the volatility, find the signal. The signal is clear: centralised regulation will not arrive in time. The signal is that the prisoner’s dilemma can be solved by a smart contract. The signal is that we have the tools. We just need to use them.
Build in public, live in truth. Let’s start now.
