Editorial illustration of three abstract geometric towers facing one another, their tops joined by a horizontal baseline, symbolizing three frontier AI labs coordinating on safety standards
Three frontier labs, one table (AI-generated illustration), AI-generated illustration, not a news photo

What happened

On September 15, OpenAI global policy head Chris Lehane confirmed at a Washington briefing that OpenAI, Anthropic and Google DeepMind have been coordinating on AI safety for weeks. According to Bloomberg, the three labs are discussing more than a vague industry initiative: a professional regulator, third-party evaluation mechanisms, and quantifiable safety baselines, modeled on the U.S. Financial Industry Regulatory Authority (FINRA). The idea traces back to a July proposal from Google DeepMind CEO Demis Hassabis.

Lehane said OpenAI does not believe the coordination needs an antitrust exemption. Separately, some U.S. lawmakers have proposed a narrow antitrust exemption that would let competitors share specific safety information.

The pushback

The FTC pushed back in public, saying the effort looks like digging a moat. The concern has precedent: the line between sharing safety information and coordinating on pricing or market division is the murkiest territory in antitrust enforcement.

The same day, OpenAI CEO Sam Altman told an all-hands meeting he supports an industry testing and auditing body, but that major labs should proceed without waiting on the U.S. government. He reiterated on September 16 that the industry is capable of developing AI safely without major harm.

Why it matters

Three labs that compete on talent, funding and release cadence are, for the first time, sitting at the same table over the mechanics of safety governance. The closest prior precedent, the Frontier Model Forum from July 2023, mostly published best practices; it never touched who evaluates, how standards are set, or what happens on violation. If the current discussions land — third-party evaluation plus quantifiable baselines — they would change how each lab self-reviews before shipping a model.

A caveat: what has been confirmed so far is discussion, not agreement. Public statements remain separate commitments; there is no formal pact. Between a proposal and a functioning body sit antitrust review and the details of enforcement.

[1][2]