The most interesting part of this story isn’t that OpenAI, Anthropic, and Google DeepMind are talking about AI safety. It’s that they’ve apparently been doing it for weeks, under the radar, before anyone confirmed it publicly. According to TechCrunch, OpenAI’s global policy chief Chris Lehane confirmed on Tuesday that the three companies have been in active discussions, corroborating earlier hints from Sam Altman and a separate report from The Information that the trio is working to build a formal standards body for the AI industry.
The backdrop here matters. Anthropic CEO Dario Amodei published an essay last Saturday calling on the AI industry to voluntarily slow frontier development and coordinate on catastrophic risk. It was a notable move from someone running one of the most capable labs in the world. Altman, Google’s Demis Hassabis, and even Elon Musk publicly backed it. Altman went further, saying OpenAI would follow Anthropic’s lead in embedding third-party evaluators inside the company to monitor safety. That’s a real commitment, not just a signature on a letter.
But the coordination raises a legitimate legal question. Altman himself acknowledged that talks between competing labs could trigger antitrust scrutiny if regulators decide the arrangement suppresses competition. Amodei’s essay proposed a narrow government waiver to allow exactly this kind of safety coordination. Lehane, however, reportedly said the companies don’t believe they need one. That’s a confident position, and it may hold, but it’s the kind of thing that looks different once a regulator starts asking questions.
The political context cuts the other way entirely. President Trump has dismissed AI safety concerns as overblown and framed any slowdown as a gift to China. His AI advisor David Sacks, who has personal financial stakes across much of the industry, has echoed that line. So the labs are trying to build a safety coordination structure without government backing, and in some cases against the active resistance of the current administration.
This is where Hassabis’s July call for a U.S. standards body with real enforcement power starts to look prescient. If Washington won’t build that infrastructure, the labs are apparently willing to try building it themselves. Whether that’s responsible self-governance or three dominant players writing their own rules is a question worth taking seriously.
At Tuesday’s meeting, Lehane also said OpenAI supports a provision in the FRONTIER Act that would require top labs to allow independent verification organizations inside their operations. That’s a concrete policy position, and it signals that at least some of the safety talk is moving toward structure rather than staying at the level of public statements.
For developers and founders, the practical stakes are real. A formal standards body, even a private one, could shape what “safe” means for model releases, what thresholds trigger slowdowns, and which organizations get credentialed to audit labs. That’s infrastructure that affects the whole industry, not just the three companies building it.
