The U.S. Wants a FINRA for AI: What That Actually Means

The Trump administration is quietly exploring something that would have seemed unthinkable just a few years ago: a federal regulator specifically tasked with pre-approving AI models before they hit the market. Treasury Secretary Scott Bessent has reportedly helped draft a proposal for an independent watchdog modeled after FINRA, the Financial Industry Regulatory Authority that polices Wall Street.

This isn't some distant policy whitepaper. According to multiple sources, the proposal is currently under review by White House Chief of Staff Susie Wiles. If it moves forward, the U.S. would join a small but growing club of nations trying to get ahead of AI risk through formal gatekeeping.

How It Would Work

The FINRA model is instructive here. FINRA isn't a government agency in the traditional sense. It's an independent, industry-funded organization that writes rules, conducts examinations, and can discipline member firms. It operates under SEC oversight but has significant autonomy in day-to-day enforcement.

The proposed AI watchdog would follow a similar blueprint. It would be funded by fees from AI companies, not taxpayer dollars. It would report to the SEC. And most controversially, it would have the power to screen, certify, or even pause the release of "frontier" AI models that exceed a certain capability threshold.

The scope here matters. We're not talking about regulating every chatbot or image generator. The target is frontier models, the bleeding-edge systems being developed by OpenAI, Google DeepMind, Anthropic, and a handful of others. These are the systems that keep safety researchers up at night.

The Surprising Alliance

What makes this proposal particularly interesting is who is supporting it. Google DeepMind CEO Demis Hassabis has publicly advocated for exactly this kind of industry-funded standards body. This is not a case of industry resisting regulation. At least one major AI lab sees value in creating a unified framework for pre-deployment testing.

The logic is straightforward. Right now, every major AI lab conducts its own safety evaluations using different methodologies and benchmarks. A FINRA-style body could standardize this process, creating a common baseline that all frontier models must meet before release. For companies already investing heavily in safety research, this levels the playing field against competitors who might be cutting corners.

But not everyone in Silicon Valley is on board. Tech trade groups have warned that heavy-handed oversight could stifle innovation, create compliance bottlenecks, and potentially drive AI development overseas. The concern isn't just about red tape. It's about whether the U.S. can maintain its competitive edge against China if American labs face regulatory hurdles that their international competitors do not.

Comparing Global Approaches

This proposal sits at an interesting intersection of regulatory philosophies. The EU's AI Act takes a broad, risk-based approach, classifying AI systems by application and imposing conformity assessments across the board. China's AI governance is even more centralized, with direct government licensing for large models.

The FINRA model offers a third way. It is binding and enforceable, unlike the voluntary frameworks that currently dominate U.S. AI policy. But it is also industry-funded and industry-influenced, avoiding the top-down command-and-control approach seen in Beijing.

Civil society groups have largely welcomed the concept but with caveats. The Center for AI Policy and similar organizations stress the need for transparent standards and public oversight. An industry-funded body risks capture if the foxes are effectively guarding the henhouse.

What Happens Next

The timeline here is deliberately vague. The proposal is in White House review. If it advances, legislative authorization would likely be required, followed by budget appropriations. Realistically, we are looking at late 2026 or 2027 before any such body could begin operations.

In the meantime, the debate will continue. Proponents argue that frontier AI poses unique risks that warrant unique safeguards. Critics counter that premature regulation could entrench incumbents, stifle open-source development, and fail to keep pace with a technology that evolves faster than any regulatory framework can adapt.

The deeper question is whether a FINRA for AI is a pragmatic solution or a category error. Financial markets are fundamentally different from AI development. Securities have clear valuation metrics, standardized reporting requirements, and mature risk models. AI capabilities are harder to measure, harder to predict, and harder to contain.

What we are really debating is whether AI safety can be operationalized the same way financial stability can. The Trump administration seems to think the answer is yes. The coming months will reveal whether Congress and the industry agree.