OpenAI in safety talks with Anthropic, Google DeepMind

The disclosure came at a delicate moment for OpenAI, which is coordinating with the two competitors it most often measures itself against. Lehane's framing — that safety cooperation is worthwhile and that no antitrust exemption is required — is a bid to head off the charge that competitors aligning on standards is simply a restraint of trade.

That charge is already in the air. On the same day, FTC chair Andrew Ferguson said he would view any exemption request from AI companies with deep suspicion, describing it as a moat built to keep challengers out. Congress is moving in two directions at once: Senator Bernie Sanders said he will introduce a bill next week to ban superintelligence, while the Banks-Schiff measure would grant a narrow exemption for sharing information on loss of control and cyber or biological threats. Bloomberg's reporting suggests the legislative route is unlikely to close this year, partly because the House leaves for the November election.

Nvidia CEO Jensen Huang sits on the other side of the argument, saying market forces are enough and that safe and rapid development can coexist. Where this lands hinges on whether regulators read cross-company safety talks as public-interest coordination or as a moat — and on whether any of the competing bills survives the election calendar.