Defensibility check: this is confirmed, on-record reporting (Lehane’s own admission, corroborated independently by TechCrunch/Reuters/Politico) — I’m keeping the antitrust angle, the FRONTIER Act detail, and Amodei’s essay clearly separated as distinct, individually-sourced threads rather than blending them into settled policy.
On Tuesday, September 15, 2026, OpenAI’s global policy chief Chris Lehane confirmed something that had been hinted at for weeks: OpenAI, Anthropic, and Google DeepMind have been quietly coordinating on AI safety. The admission, first reported by Bloomberg and independently corroborated by TechCrunch and Reuters, is a genuinely notable governance moment — but the interesting part isn’t that it happened. It’s what it does and doesn’t actually change.
What was actually confirmed
Lehane, speaking to reporters while in Washington working with lawmakers on catastrophic AI risk, said the three companies have been in safety talks for “several weeks.” This follows a chain of public signals:
- September 12: Anthropic CEO Dario Amodei published an essay calling for the industry to voluntarily slow the pace of frontier AI development to manage catastrophic risk.
- Days later: OpenAI’s Sam Altman, Google’s Demis Hassabis, and even SpaceXAI’s Elon Musk publicly supported Amodei’s call, with Altman committing OpenAI to embedding third-party safety evaluators, matching Anthropic’s move.
- The same week: The Information reported the three companies were working together on a possible AI industry standards body — something Altman reportedly told staff would need to happen without US government backing.
- September 15: Lehane’s on-record confirmation that the safety talks are real and have been ongoing.
The part that actually matters: the antitrust tension
Some industry figures, including Altman himself, have flagged that this kind of cross-company coordination could risk violating antitrust law if regulators viewed it as suppressing competition. Amodei’s original essay proposed a narrow government waiver specifically to allow this kind of safety coordination without antitrust exposure.
Lehane’s response to that, reported the same day: OpenAI doesn’t think it needs a waiver. That’s a meaningful detail — it means OpenAI is proceeding with informal, voluntary coordination without waiting for any legal cover, which cuts both ways. It could mean the talks are genuinely narrow enough to not raise antitrust concerns. It could also mean nobody has actually tested that assumption yet.
The separate, more concrete thread: FRONTIER Act
At the same Tuesday meeting, Lehane said OpenAI supports a provision in the bipartisan FRONTIER Act that would require top frontier labs to allow “independent verification organizations” into their companies to check that models are developed safely. This is a distinct, more concrete policy thread from the informal safety talks — it’s an actual legislative proposal with defined verification-organization language, not a voluntary industry conversation. Worth tracking separately, since it’s the piece most likely to become binding if it moves.
What hasn’t changed
Despite the confirmation, there’s no new binding standard, no published safety framework the three companies have jointly agreed to, and no enforcement mechanism. This is still, as of this writing, informal coordination — the kind that could evaporate, formalize into something real, or anything in between. Meanwhile, the Trump administration has publicly dismissed AI safety concerns as overblown, with AI advisor David Sacks (who holds personal financial stakes across much of the industry) echoing that framing — meaning any government-backed push toward binding oversight faces active political headwinds, not a clear runway.
The governance question that actually matters
Voluntary industry self-coordination is not a substitute for external, verifiable oversight — and the companies’ own framing implicitly concedes this. If OpenAI, Anthropic, and Google genuinely believed voluntary talks were sufficient, there’d be no need for Amodei’s antitrust-waiver proposal, no need for embedded third-party evaluators, and no need for OpenAI to publicly back a legislative provision requiring outside verification organizations. The fact that all three of those exist simultaneously suggests the companies themselves don’t think informal coordination alone gets the job done.
What actually changes for people running agents
Short answer: nothing yet. There’s no new compliance requirement, no published joint standard to adopt, and no change to how any of these companies’ models behave today. This is a signal worth watching — particularly whether the FRONTIER Act’s verification-organization provision advances — not something that requires action right now.
Sources: TechCrunch, “OpenAI, Anthropic, Google have been in talks on AI safety for weeks” (September 15, 2026); Reuters corroborating wire report; TechCrunch’s coverage of Amodei’s “Pace the Frontier” essay (September 12, 2026); Politico’s reporting on OpenAI’s FRONTIER Act position.