A Supervised SRO for AI
This idea is often referred to as a “FINRA for AI,” but that label implies it will be a copy-paste of that institution. The more accurate description of the idea is a “supervised SRO for AI,” where SRO refers to self-regulatory organization constituted by industry members. Supervised SROs are used all over finance—FINRA, NYSE, NASDAQ, DTCC, etc.—and in other industries, like NERC in energy.
The Proposal (in brief)
Membership
Every developer maintaining a certain amount of compute (rented or owned) for AI training or inference must join the SRO.
Binding Rules
The SRO sets safety and evaluation standards, and members must comply with them in order to operate.
Government Supervision
No rule takes effect without federal sign-off, and the supervising agency can compel the body to adopt, amend, or repeal a rule.
Enforcement
The SRO audits member practices for compliance with the rules and evaluates member models for excessive risk. Penalties for noncompliance include model suspension.
Funding
Assessments on compute fund the SRO, and the budget is sized to hire expert staff with salaries somewhat competitive with industry.
Notable publications
A chronology of the serious writing on a supervised SRO for frontier AI, kept current. Corrections and additions welcome.
Apr. 29, 2026 — Mark Thomas, “AI Companies Can't Regulate Themselves. They Should Regulate Each Other.” Lawfare.
Jun. 25, 2026 — Google, “A Pragmatic Approach to AI Governance in America.”
Jul. 10, 2026 — Andy Jung, “Is Private AI Regulation Constitutional?” WSJ.
Jul. 14, 2026 — Demis Hassabis, “A Framework for Frontier AI and the Dawning of a New Age”. Substack.
Jul. 19, 2026 — James Broughel, “Does the AI Industry Need Its Own FINRA?” Forbes.
Jul. 21, 2026 — Jeremy Kahn, “A FINRA for AI? The idea from Google DeepMind CEO Demis Hassabis is gaining momentum. But is it any good?” Fortune.
Jul. 23, 2026 — Vinh Nguyen, Elham Tabassi & Kat Duffy, “The U.S. Is About to Design an AI Regulator. Here's How to Get It Right.” Council on Foreign Relations.
Jul. 30, 2026 — Mark Thomas, “Designing a FINRA for Frontier AI.” Lawfare.
Aug. 24, 2026 — Tyler Cowen, “The Least Bad Way to Regulate AI.” The Free Press.
Sep. 3, 2026 — Joseph Hoefer, “Trump's FINRA for AI sounds odd, but it may be our best shot.” Washington Examiner.