Last September, the Washington Examiner published an abolitionist case against the Financial Industry Regulatory Authority and the other self-regulatory organizations Congress has empowered. It argued that SROs are an “administrative state inside the administrative state”: opaque, unaccountable private bodies exercising powers that belong in government.
Less than a year later, the Trump administration is reportedly considering a FINRA-style body for frontier artificial intelligence.
At first glance, that sounds like an odd fit, but it may be the right one.
Frontier AI moves on a different clock than conventional federal rulemaking. The National Institute of Standards and Technology’s Center for AI Standards and Innovation says frontier assessment requires “large-scale, quick-turnaround, high-signal evaluations” and continually evolving methodologies. Government can build real technical capacity, but private labs recruit on market terms, and capabilities can change faster than conventional rulemaking.
A FINRA-style model offers a way to close some of that gap. An industry-funded SRO can finance evaluators, testing infrastructure, and technical work through member assessments rather than annual appropriations. FINRA itself is funded by member firms, not taxpayers. It can update benchmarks as the technology changes, while the government retains the final say over what becomes binding.
And regulation won’t stand still if Washington does nothing. Colorado enacted a broad AI framework in 2024, then rewrote it this year and separately added requirements for conversational AI services. A national company serving Colorado has to follow each turn, and other states will make their own choices.
That doesn’t mean every AI company belongs in a new private regulator. Any federal SRO should cover only developers of genuinely frontier models that cross a clearly defined capability or compute threshold, not ordinary deployers, startups using someone else’s model, or researchers experimenting with open systems.
But a narrow perimeter creates a second problem: capture. FINRA oversees more than 3,200 member firms. A frontier-AI SRO could begin with only a handful of labs, which makes questions about capture and threshold-setting especially important. That’s why members shouldn’t control the perimeter, and why federal review has to be load-bearing rather than nominal.
FINRA’s history offers another lesson. Congress didn’t generally require broker-dealers to join an SRO until 1983. AI is still on the earlier side of that line. The administration can begin voluntarily, build the evaluation machinery, and learn what works. If Congress later makes participation compulsory, that’s the constitutional hinge.
The principle should be simple: delegate expertise, not sovereignty.
A private AI body can test models, develop benchmarks, investigate failures, and recommend action. But it shouldn’t have the final word on whether a company may build, release, or sell a model in the U.S. Congress should name a federal supervisor — commerce, supported by NIST and CAISI, is one logical candidate — with power to approve, reject, or rewrite binding rules and review serious sanctions before they take effect.
The constitutional fight over FINRA points the same way. In 2024, the D.C. Circuit objected to a system in which a FINRA expulsion could take effect before Securities and Exchange Commission review on the merits. FINRA changed its rules, and this April, a federal district court rejected Alpine Securities’ broader challenge. The lesson is straightforward: meaningful government review isn’t paperwork around the edges. It’s what keeps private expertise from becoming private government.
OPINION: TRUMP’S AI MANDATE: DEFINING KILLER ROBOTS WON’T SAVE US — HARDCODING HUMAN CONTROL WILL
Congress should also add transparency, due process, regular reporting, and an expiration date. Give any mandatory AI self-regulatory regime five years. Make it prove that it improves security and clarity without becoming a cartel for the largest labs or accumulating powers Congress never intended. Then make lawmakers vote again.
The White House is right to look for something more technical and less cumbersome than another federal commission. The conservative bargain is straightforward: delegate expertise, not sovereignty. Keep the expertise private. Keep sovereign power public.
Joseph Hoefer is a principal and chief artificial intelligence officer at Monument Advocacy.