OpenAI's global head of policy, Chris Lehane, said in Washington on Tuesday that the company has been talking with Anthropic and Google DeepMind about coordinating safety efforts across frontier models. According to TechCrunch, citing earlier reporting by Bloomberg, the talks have been running for several weeks and include the possibility of an industry standards body, still without a name, timeline, or settled design.
The context: Saturday's essay
The disclosure did not come out of nowhere. It landed three days after "We Must Pace the Frontier," the essay in which Anthropic CEO Dario Amodei argues for deliberately slowing the pace of AI capability gains — a story we covered earlier this week. OpenAI's Sam Altman and Google DeepMind's Demis Hassabis publicly endorsed Amodei's essay in the following days. Altman went further, promising that OpenAI would also embed independent external evaluators into its own processes, echoing the commitment Anthropic had made weeks earlier.
No request for an antitrust waiver
Lehane said during the briefing that the companies do not need a regulatory waiver to discuss safety. He added that OpenAI supports a provision in the FRONTIER Act that would require top frontier labs to let independent verification organizations into their operations.
It is worth separating fact from intent here. What is concrete, for now, is Altman's commitment to external evaluators and OpenAI's declared support for a legal audit requirement. The standards body itself, and the exact shape of coordination between the three companies, remain open questions.
What it means for people building on these models
For engineering teams running GPT, Claude, or Gemini in production, the announcement matters less as news and more as precedent: three direct competitors, who normally fight over every benchmark percentage point, are publicly admitting they talk to each other about shared safety limits. If that turns into an actual common audit standard, instead of three incompatible proprietary frameworks, teams certifying AI products in regulated industries could end up with more predictable risk-assessment processes ahead.
What is still missing is what is always missing from this kind of joint announcement: a timeline. Reports that the three labs have been talking for weeks are very different from the labs publishing the same technical standard. Worth checking back in a few months to see if anything survives beyond coordinated press statements.