OpenAI, Anthropic and Google coordinate on AI safety without antitrust waiver

This is one of the clearest public signals yet that the leading frontier labs are trying to coordinate on safety while staying on the right side of antitrust law. Lehane’s message is essentially that OpenAI can talk with Anthropic and Google DeepMind about catastrophic risk mitigation without entering into an explicit output‑ or investment‑limiting cartel. If true, that creates room for shared safety baselines, incident‑response protocols and perhaps joint red‑teaming, even as the companies keep competing on capabilities.

For the race to AGI, the significance is double‑edged. On one hand, safety coordination can help reduce correlated failure risks: if all three labs quietly adopt similar guardrails on things like model‑enabled bio threats or offensive cyber operations, the probability of a worst‑case outcome falls. On the other hand, the same channels can accelerate the diffusion of best practices and techniques that make it easier to field ever more capable systems quickly once a safety box is ticked.

The political backdrop matters too. With US antitrust enforcers signaling skepticism toward formal exemptions, labs may lean into softer, disclosure‑based coordination and industry standards instead of hard caps on compute or training runs. That is more palatable to investors and boards, but leaves open the question of who, if anyone, can actually say “stop” if shared safety monitoring shows things going off the rails.