OpenAI, Anthropic, Google have been in talks on AI safety for weeks
Sep 15, 2026, 8:47 AM · TechCrunch
OpenAI’s Chris Lehane told reporters the company has been working with Anthropic and Google DeepMind on AI safety for weeks—amid antitrust worries and Amodei’s call for a narrow coordination waiver.
Why it matters
Rebecca Bellan at TechCrunch reports OpenAI global policy chief Chris Lehane saying OpenAI has been in AI-safety talks with Anthropic and Google DeepMind for weeks (also reported by Bloomberg). Lehane was in Washington on catastrophic-risk issues after Dario Amodei’s Saturday essay urging slower frontier pacing and embedded evaluators—moves Altman, Hassabis, and Musk publicly backed in varying degrees.
Altman and others have flagged antitrust risk if coordination suppresses competition; Amodei proposed a narrow government waiver for safety conversations.
Private talks plus public letters is the governance story of the week.
From the desk
We’re glad rivals are talking—and we want sunlight on what they’re allowed to agree.
Weeks of OpenAI–Anthropic–DeepMind safety discussions, confirmed by Lehane, make Amodei’s essay look less like a solo op-ed and more like the public face of an ongoing channel. Useful if the channel produces shared eval access, incident norms, and deployment thresholds. Dangerous if it becomes a quiet cartel on pricing or talent dressed up as safety—or if fear of antitrust becomes an excuse to never commit.
We’re for a narrow, transparent waiver that enables safety coordination without blessing market allocation. The harm of opaque talks: public trust collapses when “safety” looks like collusion.
I’m watching whether Washington drafts that waiver, and whether embedded evaluators actually get desks—not only press quotes.
Context
Amodei’s essay ran the prior Saturday; Altman pledged to match embedded third-party evaluators. Lehane’s comments came Tuesday, September 15, 2026, per TechCrunch.
Who feels it
- U.S. lawmakers and DOJ/FTC
- Clarify what safety coordination is permitted; consider Amodei’s narrow-waiver idea explicitly.
- Frontier labs
- Convert talks into dated operational commitments—evaluators, reporting, rate limits.
- Public and civil society
- Demand summaries of what was discussed; secrecy feeds capture narratives.
What to watch
- Any formal antitrust guidance or waiver language from U.S. agencies.
- Named evaluator orgs seating inside OpenAI/Anthropic/Google.
- Joint safety standards documents emerging from the talks.