The heads of the leading frontier AI companies have coalesced around an unusual idea: a shared, industry-funded safety regulator modeled on FINRA, the private body that oversees US broker-dealers. The proposal, advanced in a September 20 document, would give the sector a common framework for testing, incident reporting and independent audits before governments impose one.
The FINRA analogy is deliberate. Like financial markets, frontier AI development concentrates risk in a handful of well-capitalized players whose failures spill over onto the public. A self-regulatory organization would sit between voluntary commitments and full statutory regulation, with the power to set standards members must meet to ship models.
Amodei, Altman and the Antitrust Wrinkle
The regulatory push arrives as the industry's own leaders are speaking more urgently about risk. Anthropic chief executive Dario Amodei wrote in a September 12 essay that the industry is moving too fast relative to its ability to understand what it is building, and called for external oversight with real leverage. OpenAI's Sam Altman told audiences at Dreamforce on September 16 that "the world is right to be afraid" of advanced AI, a striking departure from his earlier optimism.
The cooperation has a legal complication. On September 19, an antitrust complaint filed in the Northern District of California alleged that major labs coordinated on a collective slowdown of model releases, restraining competition under the banner of safety. The suit argues that safety framing cannot immunize what is, in substance, an agreement among rivals.
Attorney Ryan Lehane, commenting on the tension between collaboration and competition law, said "it is better to try to work together to prioritise safety," but noted that regulators will want visible guardrails around what labs discuss when they meet.
Whether the FINRA-style body survives legal scrutiny, wins government backing or collapses under antitrust pressure will be one of the defining technology-policy fights of the next year.
