Entropy wins. Always check the fees. But the fees Scott Bessent proposes aren’t gas fees. They’re compliance costs. And they’re about to reshape the AI landscape in ways most developers haven’t modeled yet.
Context
Treasury Secretary Scott Bessent floated a proposal: create an independent agency modeled after FINRA to regulate frontier AI models. The agency would sit under the SEC. Its mission: enforce compliance for models deemed “frontier.” The logic is clear. AI systemic risk is financial systemic risk. Same playbook. Same institution. The Financial Industry Regulatory Authority polices broker-dealers. Bessent wants a similar body to police AI developers.
Core
Let’s audit the technical assumptions. The proposal defines “frontier” by compute thresholds. Likely 10^26 FLOPs or above. That sounds precise. It’s not. Compute thresholds shift with hardware. A model trained today on 10^27 FLOPs may be obsolete tomorrow. More importantly, the threshold creates a binary gate. Pass it, and you face full audits. Stay below, and you’re free. This invites gaming.
I saw the same pattern in DeFi. Liquidity mining APY is subsidized TVL. Here, “frontier” classification will be gamed. Developers will train models at 9.9^26 FLOPs to avoid the gate. The models will be just as dangerous. The math doesn’t lie. The incentive structure does.
Now, the enforcement mechanism. The agency will require developers to submit safety audits. Model red-teaming, bias testing, adversarial robustness scores. On paper, this looks rigorous. In practice, it’s tick-box compliance. Based on my experience auditing Solidity in 2017, I learned that standard audit checklists miss the edge cases. The real vulnerabilities live in the assumptions. The same holds here.
The proposal assumes safety can be quantified and verified by a third party. It cannot. Safety is a moving target. A model released today passes all tests. Tomorrow, a jailbreak emerges. The agency’s certification becomes a false flag. Investors will treat the “SEC-approved” label as a guarantee. It’s not. It’s a liability.
Contrarian
The blind spot is regulatory capture. Bessent’s agency will be staffed by lawyers and ex-FINRA officials. Not cryptographers. Not ML researchers. The people who write the rules will not understand the systems they regulate. This gap produces two outcomes.
First, compliance will become a cost barrier. Large incumbents—OpenAI, Google, Microsoft—already have legal teams and lobbying budgets. They will shape the rules to favor their architectures. Open-source projects like Mistral or DeepSeek cannot hire a Washington law firm. They will be forced to stay below the compute threshold or exit the US market entirely. The result is a de facto monopoly on frontier AI.
Second, the agency will become a bottleneck. Every model update requires re-certification. Every new training run triggers a review. The timeline for approval will stretch from weeks to months. Innovation slows. The US loses its edge. 2017 vibes. Proceed with skepticism.
There’s a deeper risk: the politicization of “safety.” The SEC is not a neutral arbiter. Its leadership changes with administrations. A future Secretary could redefine “frontier” to include models that critique government policy. The agency becomes a censorship tool. This is not hypothetical. The history of financial regulation shows how rules meant to protect markets are weaponized against competitors.
Takeaway
Bessent’s proposal is a land grab. It locks in the current oligopoly and strangles the next generation of builders. The real vulnerability is not model alignment. It’s regulatory alignment with incumbents. Impermanent loss is real. Do your math. The cost of compliance will devour startup runway. The only winners are the gatekeepers. And entropy. Entropy always wins. Check the fees. They’re coming.
