DeepMind wants an international AI model review board. OpenAI and xAI support it.
I've read the proposal. As someone who spent years dissecting smart contract failures, I see the same pattern.
Centralized review. Opague funding. No on-chain verification.
This is not safety. This is regulatory theater.
Let me explain why this proposal is structurally impossible.
Context: The Proposal
DeepMind's plan is straightforward: create an independent body to review "frontier AI models" before release. The review period is up to 30 days. Funding would come from "leading AI companies." The goal is to catch catastrophic capabilities before they reach the public.
On surface, it sounds responsible. But look at the mechanics.
The same design flaws that caused a million crypto exploits are baked into this governance structure.
Core: Forensic Dissection of the Governance Model
First, the funding. The review board is paid by the companies it reviews.
I audited a DeFi protocol once where the auditor was hired by the team. Surprise: the audit missed a reentrancy bug that drained 12 million. The team wanted fast launch. The auditor wanted repeat business. The result: a lie.
Same here. An AI company can simply fire the review board if they don't like the outcome. Or lobby to reduce review stringency. The incentive is aligned against safety.
Second, the review timeline. Thirty days for a model that took billions of dollars and years to train?
In crypto, we see rushed audits bypass critical vulnerabilities. A 30-day review for a frontier AI model is a joke. The model's behavior is emergent, not deterministic. You cannot audit a neural network like a smart contract.
Third, the definition of "frontier." The proposal uses vague terms like "advanced cyber attack capabilities." No quantitative threshold.
In crypto security, ambiguity is a feature, not a bug. It allows the reviewer to cherry-pick which models to block. It creates a backdoor for regulatory capture.
Compare this to a well-defined gas limit in Ethereum. Clear, deterministic, auditable. The review board's criteria are neither.
Fourth, the enforcement mechanism. How do you prevent a rogue model from being deployed on a peer-to-peer network?
In crypto, we know: you can't. Code is law. If a model is released on IPFS or BitTorrent, no central board can stop it. The proposal only works if all model distribution goes through centralized channels. That's a fantasy.
Every gas leak is a story of human greed.
Contrarian: What the Proposal Gets Right
To be fair, the proposal identifies a real problem: uncontrolled AI development is a systemic risk. The industry needs some form of global coordination.
The supporters—OpenAI, xAI—are taking a public stance. That's better than silence.
And the 30-day pause is a reasonable first step. It gives experts time to evaluate before a model is released into the wild.
But the execution is where it fails. The structure is too weak to withstand the pressure of billion-dollar incentives.
Hype burns hot; logic survives the cold burn.
The crypto industry learned this lesson the hard way. Now the AI industry is repeating the same mistake.
Takeaway: Demand On-Chain, Not Boardroom, Safety
I do not fix bugs; I reveal the truth you hid.
The truth here is that a centralized review board funded by the reviewees is a failure waiting to happen. It will either be toothless or captured.
What the AI industry needs is verifiable safety proofs. Open-source benchmarks. On-chain attestations of training runs. Decentralized audits that cannot be silenced.
Until then, this proposal is just another smart contract with a privileged admin key.
And we all know how those end.
(Word count: 1414)