Story
September 15, 2026
AI’s biggest labs are talking safety—but coordination brings its own risks
OpenAI, Anthropic and Google DeepMind appear to agree that frontier AI risks demand collective action, but they face a basic contradiction: the cooperation they say is necessary for safety could itself draw legal scrutiny and political resistance.
OpenAI’s global policy chief, Chris Lehane, said Tuesday that the company had been in AI-safety discussions with Anthropic and Google DeepMind for weeks—a confirmation that the industry’s fiercest competitors are exploring common ground on catastrophic-risk safeguards.1
The talks followed Anthropic chief executive Dario Amodei’s weekend call for frontier labs to work together to slow development when necessary and avert extreme harms. OpenAI CEO Sam Altman backed the broader push, saying his company would join Anthropic in bringing in third-party evaluators to monitor safety, according to the report.1
That is a significant shift in tone from an industry built on speed. The reported agenda includes coordinated responses to dangerous capabilities and, separately, work toward an AI standards body. Lehane, in Washington to engage lawmakers on catastrophic AI risks, said OpenAI supports a FRONTIER Act provision that would give independent verification organizations access to leading labs.1
Yet the proposed cooperation comes with a catch. Altman and others have acknowledged that coordination among rivals could run into antitrust law if it is judged to suppress competition. Amodei has proposed a narrow government waiver for safety cooperation; Lehane’s position is that the companies do not need one.1
The debate is playing out against a sharper political divide. President Trump has dismissed AI-safety concerns and argued that slowing US development would advantage China, while adviser David Sacks has called existential-risk fears overstated.1
Outside the companies, the reaction ranges from skepticism to urgency. Yann LeCun amplified a post alleging Anthropic has a dense network of ties to catastrophic-risk funders—an allegation in the repost, not independently established by the material provided.
2 Musk, meanwhile, shared a summary of remarks stressing that “AI can be very dangerous.”
3 In another repost, he highlighted a “practical framework” for making frontier AI safer, centered on the proximity of OpenAI’s and Anthropic’s capabilities.4
The emerging consensus, then, is not that the labs should stop competing. It is that they may need rules for the moments when competition becomes too risky to manage alone.