Can AI Companies Police the Frontier Without Becoming a Cartel?
The companies are exploring a self-regulatory organization modelled on the Financial Industry Regulatory Authority, which oversees US brokers and investment firms.[1] OpenAI said discussions had been underway for several weeks and that it did not believe the competitors needed an antitrust waiver to coordinate on safety.[5] The wider proposal combines embedded third-party evaluators, common standards among frontier labs and eventual international coordination.[9]
No leading AI company has yet implemented a sustained, comprehensive slowdown, making enforceable testing, auditing and incident-reporting rules the practical test of whether “pacing” changes development behavior.[6] Critics also warn that rules designed by dominant labs could burden smaller competitors, restrict open-source development or substitute voluntary promises for public regulation.[8]
Key insights
- The proposed approach distinguishes pacing from stopping: Sam Altman said progress should continue, but more slowly than it otherwise would.[9]
- Independent evaluators could be embedded inside labs to verify safety commitments, report incidents and flag problematic findings.[8][9]
- A measurable agreement could include auditor access to companies’ compute budgets, allowing outsiders to assess whether promised slowdowns are real.[8]
- Congressional proposals range from mandatory harm mitigation and emergency shutdown capabilities to risk reports for the Commerce Department and a pause on advanced development pending federal safety rules.[6]