OpenAI, Anthropic and Google Confirm Weeks of Secret AI-Safety Talks

The three frontier labs have been coordinating on safety — and now face the antitrust question their own coordination raises.

Published: 2026-09-16 Category: Quick Take Sources: TechCrunch — OpenAI, Anthropic, Google have been in talks on AI safety for weeks

What happened

OpenAI's global policy chief, Chris Lehane, told reporters on Tuesday that the company has been working with rivals Anthropic and Google DeepMind on AI safety for weeks — as first reported by Bloomberg. Lehane is reportedly in Washington to work with U.S. lawmakers on the catastrophic risks associated with AI.

The revelation follows Anthropic CEO Dario Amodei's essay, published Saturday, calling on the industry to work together to slow the pace of frontier AI and avoid catastrophic risks. AI leaders across the industry — including OpenAI's Sam Altman, Google's Demis Hassabis, and SpaceXAI's Elon Musk — supported the call, with Altman saying OpenAI would embed third-party evaluators to monitor for safety, matching Anthropic's approach. Lehane confirmed the three companies have held weeks of safety talks but reportedly said they don't need the narrow government waiver Amodei proposed to allow such coordination.

The news backs up earlier hints. Altman told Fortune late last week that he had been in private discussions with other AI leaders, and The Information reported the three companies are working to create a standards body for the AI industry — something Altman reportedly told staff would need to happen without U.S. government support. Hassabis separately called on the U.S. in July to establish a standards body with the power to screen the world's most advanced models. President Trump, for his part, has dismissed safety concerns as a hoax and pushed back on tighter regulation.

The coordination paradox

Here is the tension at the heart of this story: the same people who say frontier AI is so dangerous that the industry must coordinate are now coordinating — and in doing so are brushing against antitrust law. If competitor coordination suppresses competition, it can be illegal. Amodei's essay anticipated exactly this, proposing a narrow government waiver to legalize safety-focused coordination. Lehane's reported position — that the firms don't need one — is bolder, effectively arguing that voluntary safety coordination among rivals is fine without explicit legal cover.

That matters because it's the difference between a defensive posture and an institutional one. A waiver would have given the talks formal legitimacy and a boundary — a legal definition of what kind of coordination is allowed. Skipping the waiver leaves the coordination in a gray zone: happening de facto, justified by catastrophic-risk rhetoric, but without a clear line separating legitimate safety work from potentially illegal collusion on pace or output. The labs are effectively building shared infrastructure (evaluators, a standards body) that sits closer to the center of their business than ordinary safety research does.

Why this matters

The fact that the three biggest Western frontier labs are privately coordinating on safety is significant regardless of the antitrust question. It's a de facto recognition — from the people best positioned to know — that no single lab can unilaterally manage frontier risk, and that uncoordinated racing is itself a hazard. Amodei's essay made the slowdown case in public; these talks are the private, operational version of that argument.

The institutional question is where it gets sticky. Creating a shared standards body and embedding third-party evaluators across rival companies is, in effect, building a private-sector regulator — governance that looks a lot like public authority but is being constructed without a public mandate or legal framework. Some, including Hassabis, want that power vested in a government body; Altman reportedly wants it built without government support. The coming fight is about whether AI's guardrails get set by a public institution accountable to citizens, or by a private standards body accountable to the labs themselves. Lehane's "we don't need a waiver" comment suggests the labs want to keep that question as far from elected officials as possible.

Reporting via TechCrunch's "OpenAI, Anthropic, Google have been in talks on AI safety for weeks" (September 15, 2026).