OpenAI safety talks draw Anthropic, Google DeepMind rivals

OpenAI safety talks with two top rivals are advancing as regulators question whether coordination could narrow competition.

Chris Lehane, OpenAI’s global policy chief, said Tuesday in Washington that engagement with Anthropic PBC and Google DeepMind had been under way for several weeks before his remarks. He said OpenAI does not believe the three AI firms need an antitrust waiver to coordinate on safety matters.

Three labs test coordination

The discussions place three of the most closely watched AI developers in a delicate position: they are competing to build more capable systems while also trying to show policymakers they can manage risks. Google DeepMind is part of Alphabet Inc., giving the talks a direct link to one of the largest technology balance sheets in the sector.

Lehane framed the effort as a safety priority rather than a market arrangement. “It’s better to try to work together to prioritize safety,” he said at the briefing, a formulation that leaves open what the firms would share, publish or withhold.

Amodei essay shifts tempo

The timing follows a 3,800-word essay posted Saturday by Anthropic Chief Executive Officer Dario Amodei, who urged restraint in developing the most advanced AI systems until researchers better understand the threats. His argument helped push worries about economic disruption and security exposure further into the mainstream policy debate.

The essay drew support from OpenAI CEO Sam Altman and SpaceXAI chief Elon Musk, aligning three prominent AI figures around a slower-development message even as their companies remain rivals. That alignment is politically useful for safety advocates, but it also gives regulators a reason to examine whether joint action would harden incumbent advantages.

Ferguson questions AI moats

Federal Trade Commission Chair Andrew Ferguson said Tuesday he would be “deeply suspicious” of any request by AI companies for an exemption allowing coordination. Ferguson said he viewed the companies as “asking for barriers to entry” that could protect them from challengers, adding: “That sure sounds like moat digging.”

His comments put OpenAI’s safety case against a familiar antitrust concern: dominant firms can use technical rules, compliance demands or shared restraint to slow smaller competitors. For startups, the key issue is whether any safety framework becomes a practical cost of entry rather than a narrow risk-management exercise.

If the talks remain limited to technical safety practices, OpenAI could gain policy credibility without asking regulators to suspend competition rules. At the macro level, that path would support broader AI adoption by reducing the chance that governments respond with blunt limits on deployment; for the wider industry, it would create pressure to document safety work more clearly.

If the effort instead moves toward formal limits on advanced model development, the antitrust risk rises. OpenAI would face closer scrutiny over whether safety coordination also protects its market position, while the sector could split between large labs able to absorb compliance costs and smaller developers arguing that the rules lock them out.

The open questions are concrete: whether Anthropic and Google DeepMind publicly confirm the scope of the work, whether the companies propose shared testing or restraint, and whether the FTC treats the effort as safety coordination or market defense. Those answers will determine whether the talks become an industry template or another front in the AI competition fight.