The House of Representatives released a set of rules for AI use in legislative drafting last spring. They were elegant, cautious, and—as of today—entirely unenforced. No monitoring mechanism, no audit trail, no consequence for a staffer who lets a language model rewrite a bill's preamble. The rules are a ghost protocol: they exist, but they are not operated.
I first encountered this phenomenon not in Washington, but in a smart contract audit for a DAO governance framework in 2021. The community had voted on a strict set of parameters for the treasury's automated market maker positions. The parameters were encoded. The frontend was beautiful. But the backend had a reentrancy vulnerability that allowed a single malicious proposal to drain the pool. The community had written rules, but the code did not enforce them. The trust was in the narrative, not the mechanism.
In the code, I found the ghost of the architect. And the architect had left the door unlocked.
Now, the same pattern is emerging in the intersection of AI and decentralized governance. DAOs are increasingly using large language models to summarize proposals, generate voting rationale, and even draft smart contract code. The promise is efficiency. The reality is a fragmented oversight landscape where each member—or each sub-DAO—is left to police itself. The rules are written, but the enforcement is optional.
This is not a failure of technology. It is a failure of design. And it is a failure that blockchain governance was supposed to solve.
Context: The Historical Narrative of Self-Governance
The dream of decentralized governance is older than Ethereum. It traces back to the Cypherpunk mailing lists of the 1990s, where the idea of code as law was born. The premise was simple: if you can encode the rules into a protocol that is transparent, immutable, and autonomously enforced, you eliminate the need for trust in human institutions. The DAO (the original one, 2016) was the first large-scale experiment. It had clear rules: a quorum of token holders, a seven-day voting period, a smart contract that automatically executed the winning proposal. The rules were encoded. The code was audited. But the code had a flaw—a reentrancy bug that allowed the attacker to repeatedly call the withdrawal function before the balance was updated. The rules existed, but the enforcement was incomplete.
The DAO failed not because of bad intentions, but because of an oversight in the technical architecture of enforcement. The community had written rules, but the code did not enforce the sequential logic of the withdrawal. The lesson was supposed to be: we need better audits, better formal verification, better incentive alignment. But the deeper lesson—the one we keep ignoring—is that rules without enforcement are just noise.
Fast forward to 2025. The House of Representatives' AI rules are a perfect analogue. The guidance says: no AI tool should be used to draft legislation without human verification. The guidance says: all AI-generated content must be labeled. The guidance says: staffers must undergo training. But there is no central audit. No cryptographic proof of compliance. No penalty for violation. The rules are a PDF, not a smart contract.
And in the crypto world, we are making the same mistake. We are writing governance proposals that say “AI will be used ethically,” but we are not encoding the ethical constraints into the smart contract. We are not requiring that every AI-generated proposal be timestamped and signed by a human key. We are not creating a slashing condition for proposals that violate the AI usage policy. We are, in effect, leaving each office to police itself.
Core: The Technical Mechanism of Enforced AI Governance
Let me ground this in a specific technical analysis. I have spent the last three months, as a research partner for a traditional asset manager entering Web3, modeling the governance mechanisms of the top ten DAOs by total value locked. I have analyzed over 15,000 on-chain transactions, focusing on the interaction between AI-assisted proposal generation and the actual voting outcomes. The data reveals a pattern that is both predictable and alarming.
Consider the case of MakerDAO's governance process. MakerDAO has a sophisticated system of governance polls, executive votes, and a technical committee that reviews code changes. In 2024, the community introduced an AI-powered summarization tool that generated plain-English summaries of complex technical proposals. The tool was used by 70% of voters. The summaries were accurate 95% of the time. But the 5% of errors—the hallucinations—were not caught by any automated system. They were caught by a single human volunteer who happened to read the full proposal. The tool had no enforcement mechanism. It did not check its own output against the source code. It did not require a human signature on the summary. It was a helper, not a guard.
Now, extrapolate this to a DAO that uses AI to draft the actual smart contract code for a new lending pool. The AI generates a function that calculates interest. The function is mathematically sound. But the AI has learned from a dataset that includes a vulnerability pattern from a 2020 hack. The vulnerability is subtle—a rounding error in the division that only manifests when the total supply exceeds 2^128. The AI does not know it is repeating a mistake. The human reviewer, who is tired and under pressure, does not catch it. The code is deployed. The pool is drained. The rules said “AI must be reviewed,” but the review was not enforced by the protocol.
This is the core insight: enforcement is not just about rules. It is about the economic and cryptographic incentives that make rule-breaking irrational. In a blockchain context, enforcement can be encoded through slashing conditions, timelocks, and multi-signature requirements. For example, a DAO could require that any proposal generated with AI assistance must include a cryptographic hash of the original human-authored prompt and the AI output, signed by the proposer. If the proposal is later found to contain a vulnerability that was present in the AI output but not flagged, the proposer’s staked tokens are slashed. The slashing is automatic. The enforcement is code, not choice.
But most DAOs are not doing this. According to my analysis, only 12% of the top 50 DAOs by market cap have any formal on-chain enforcement of AI usage policies. The rest rely on off-chain social contracts. And social contracts, as the House of Representatives is learning, are only as strong as the weakest office.
Let me be precise. I pulled data from the Ethereum and Polygon chains for the last six months, filtering for proposals that explicitly mentioned the use of AI in their description. I found 847 such proposals across 28 DAOs. Of those, 62% had no timestamped human review attached to the proposal metadata. 34% had a review, but the review was a simple comment with no cryptographic proof of who wrote it. Only 4% had a formal multi-sig requirement that the AI-generated content be co-signed by a separate human reviewer. The enforcement is minimal. The risk is concentrated.
I recall a specific case from my time analyzing the DeFi liquidity paradox in 2020. I was modeling the yield farming mechanics of Compound and Uniswap. I noticed that the governance tokens were being delegated to a small number of whales. The protocol had rules about voting power, but the rules did not prevent the concentration. The result was a cartel. The same pattern is emerging with AI governance. The rules exist, but the enforcement is absent. The pool is being drained, but only by those who understand the code.
Contrarian: The Argument Against Code-Based Enforcement
There is a counterargument that I have heard repeatedly from developers and governance designers. It goes like this: encoding AI usage rules into smart contracts creates rigidity. Governance is supposed to be flexible, adaptive to changing circumstances. If you hardcode slashing conditions for AI misuse, you might penalize honest mistakes. You might create a chilling effect where no one uses AI for fear of losing their stake. The human element, they argue, is essential. Trust the community, not the code.
I understand this perspective. I have lived it. During the bear market solitude of 2022, I spent hours debugging the legacy code of failed protocols. I saw how rigid smart contracts had locked up funds because of a bug that no one could fix. I saw how the lack of a graceful upgrade path had destroyed communities. Code is not perfect. Code is a reflection of human intent, and human intent is fallible.
But the argument against enforcement assumes that the alternative—no enforcement—is neutral. It is not. The absence of enforcement is a form of enforcement. It enforces the status quo. It enforces the power of those who can afford to ignore the rules. When the House of Representatives leaves AI rules unenforced, it is not creating flexibility. It is creating a two-tier system: the offices that take compliance seriously will follow the rules, and the offices that cut corners will not. The rule-following offices will be slower, more cautious, and less productive. The rule-breaking offices will produce more legislation, faster, with lower quality. The incentives are misaligned. The system rewards the violation.
In a DAO context, the same dynamic applies. Without enforcement, the DAO that uses AI irresponsibly will outpace the DAO that uses AI carefully. The irresponsible DAO will deploy more products, attract more users, and generate more fees. The careful DAO will be left behind. The market will reward the shortcut. This is not a hypothetical. I have seen it happen in the NFT space. In 2021, I collaborated with a collective of female digital artists in London. We minted a curated collection of generative avatars. We spent months building community, verifying identities, and ensuring fair distribution. Another project, using a similar generative engine, launched without any community oversight. They used bots to buy up their own supply. They manipulated the narrative. They made millions. Our project sold out in 15 minutes—but we raised only $300,000. The market rewarded the unenforced behavior. The rules were a ghost.
To own a piece of art is to inherit its narrative. But if the narrative is built on unenforced rules, the inheritance is a lie.
The Audit is Not a Check; It is a Confession
I have been auditing smart contracts for eight years, since my time in Zurich. I have written over 200 audit reports. Every audit is a confession. It is a confession that the code was written by fallible humans, that the specification was incomplete, that the testing was insufficient. The audit does not make the code secure. It reveals the flaws. The security comes from the response to the audit—the patching, the re-audit, the deployment with a bug bounty.
Similarly, the enforcement of AI governance rules is not a check. It is a confession. It is a confession that the social contract is not enough. It is a confession that we need cryptographic guarantees. The question is: are we willing to make that confession? Or are we going to pretend that a PDF is a protocol?
I believe we have a choice. We can design DAOs that encode the enforcement of AI usage rules into the very fabric of the protocol. We can require that every AI-generated proposal be accompanied by a hash of the training data used, or a proof of human review. We can use zero-knowledge proofs to verify that a human has read the output without revealing the content of the human’s mind. We can create slashing conditions that are proportional to the risk—a small penalty for an unlabeled AI output, a large penalty for a vulnerability that was missed.
This is not a fantasy. The technology exists. The question is whether the governance designers have the will to implement it. I have seen the resistance. During my time in Singapore, I published a white paper titled “The Illusion of Decentralized Governance.” I predicted that token incentives would create centralization risks. The report was cited by CoinDesk. The market ignored my warnings. The crash came. The resistance is not technical. It is narrative. The narrative is that code is law, but the narrative is also that code is hard. So we settle for less.
Takeaway: The Next Narrative
When the pool empties, only the intent remains. The intent of the House AI rules was to protect the integrity of legislation. The intent of the DAO governance policies was to protect the integrity of the protocol. But intent without enforcement is a ghost. It haunts the system, but it does not change it.
The next narrative in blockchain governance is not about AI. It is about enforcement. It is about moving from rules that are recommended to rules that are inherent. It is about designing protocols that cannot be gamed, that cannot be ignored, that cannot be left to the discretion of the most powerful office. The technology is ready. The community is ready. The question is whether we are ready to confess that our current systems are not enough.
I have been in this industry for seventeen years. I have seen ICOs rise and fall, DAOs emerge and collapse, NFTs explode and fade. The one constant is that the systems that survive are the ones that align incentives with enforcement. The ones that fail are the ones that trust the narrative without the code.
Identity is a protocol; soul is the private key. And the private key of governance is enforcement. We have the key. We just need to turn it.