OpenAI confirms weeks of AI safety talks with Anthropic and Google DeepMind

OpenAI has confirmed that it spent recent weeks in talks with Anthropic and Google DeepMind about AI safety, according to a TechCrunch report published on September 15.

The disclosure matters less for the details it provides than for what it confirms: the three most prominent frontier labs have been speaking with one another about risk while competing head-to-head on model capability, compute and market share.

The timing is notable. According to the same report, the current US administration's team has been dismissing safety concerns and pressing to keep pace with China. The lab-level conversations are taking place against that policy backdrop.

That creates a striking contrast: as urgency fades on the regulatory side, the companies themselves appear to be filling the gap, trying to establish some shared understanding or baseline around risk.

TechCrunch's report does not spell out the terms discussed, the seniority of the participants, or whether any written outcome emerged. What can be confirmed for now is narrower: the talks happened, and they ran for weeks.

Why the signal matters: frontier AI safety governance is developing a path driven by firms themselves. Where no unified regulatory framework exists, private coordination among major vendors can move faster than public legislation — and is correspondingly harder to scrutinize.

Competition is the other layer. The three labs are potential partners on safety and direct rivals in business. Cooperative and racing dynamics coexist, so any shift in product releases, compute spending or government relations on one side can change the trajectory of the conversation.

What to watch next: whether these discussions turn into public safety commitments, shared evaluation standards or joint statements; whether the administration responds; and whether the three coordinate beyond safety on wider policy positions. The report opens a window onto a channel that was not previously public, but many details remain undisclosed.