OpenAI, Anthropic and Google DeepMind Have Been Quietly Talking About AI Safety for Weeks
The three biggest names in AI are coordinating behind closed doors on safety standards, and now one of them has confirmed it publicly.
Key points
- OpenAI's global policy chief Chris Lehane confirmed on Tuesday that OpenAI, Anthropic and Google DeepMind have been in talks on AI safety for several weeks.
- The three companies are reportedly working together to create an industry standards body, a self-policing organisation that would set rules AI labs agree to follow.
- Anthropic CEO Dario Amodei published an essay on Saturday calling on AI companies to slow frontier AI development to avoid catastrophic risks.
- OpenAI's Sam Altman publicly supported Amodei's essay and said OpenAI would allow independent, outside evaluators into the company to check that its AI is built safely.
- President Trump has dismissed AI safety concerns as a hoax, and his AI advisor David Sacks called fears about existential risk from AI overblown.
For months, the biggest AI companies have looked like fierce competitors. Turns out they've also been having a very different kind of conversation in private.
Chris Lehane, OpenAI's head of global policy, told reporters on Tuesday that his company has spent weeks in talks with rivals Anthropic and Google DeepMind about AI safety, as first reported by Bloomberg. Lehane was in Washington to meet with lawmakers about the risks posed by the most powerful AI systems, known as frontier models.
What are they actually agreeing on?
The talks focus on safety standards and, reportedly, a shared industry body to enforce them. Altman reportedly told staff the organisation would have to exist without U.S. Government backing.
Google DeepMind's CEO Demis Hassabis had called on the U.S. Government in July to establish exactly this kind of body, one with the power to screen the world's most advanced AI models and coordinate industry-wide slowdowns if dangers grew serious enough. We've tracked his public positioning across eleven stories since July.
The timing follows Anthropic CEO Dario Amodei's essay, published Saturday, urging the industry to slow the pace of frontier AI development. Altman backed it publicly. So did Elon Musk. Altman also said OpenAI would embed third-party evaluators inside the company to monitor for safety problems.
Could this coordination break competition law?
Yes, possibly. Altman himself flagged that coordinating with rivals could risk violating antitrust law, the rules that stop companies colluding to shut out competition, if regulators decide the talks suppress competition rather than protect the public. Amodei's essay proposed a narrow government waiver to give the companies legal cover. Lehane, however, said publicly on Tuesday that no waiver is needed.
That's a significant claim, and one that will get scrutiny.
At the same Tuesday meeting, Lehane said OpenAI supports a provision in the FRONTIER Act, a proposed piece of legislation, that would require top AI labs to let "independent verification organisations" inside their companies to check that models are built safely. That lines up with what Altman already promised voluntarily. AI2Day first covered the FRONTIER Act on 14 September.
Should you worry about safety theatre?
The honest concern here is that a self-policing body run by the same companies it's supposed to police is only as credible as the people inside it. If the standards body forms with no independent membership and no government mandate, it's a press release dressed up as governance.
The White House isn't enthusiastic regardless. Trump has called AI safety a hoax, and Sacks said fears of existential risk are overblown. Any formal framework will come from the industry itself, not from Washington.
Watch whether the standards body gets announced, and whether it has any real teeth once it does.
