Story
September 16, 2026
Rivals Race to Build AI Safety Rules Before the Politics Overtake Them
OpenAI, Anthropic and Google DeepMind are discussing shared safety standards and possible oversight for frontier AI. The unusual cooperation has drawn support from industry leaders but also revived fears that safety rules could become a brake on competition.
The push began in July, when Google DeepMind chief Demis Hassabis proposed a U.S.-led AI “Standards Body” — a public-private or self-regulatory arrangement with federal oversight. The idea was to give competing frontier labs a shared mechanism for confronting risks none can credibly handle alone.1
By Saturday, Anthropic chief executive Dario Amodei had sharpened the urgency, urging companies to slow the improvement of their most advanced models to avoid catastrophic outcomes. Hassabis called the approach “the right path forward,” while OpenAI chief Sam Altman said pacing frontier systems had become a “primary topic” inside OpenAI.1
On Tuesday, OpenAI confirmed it had been talking with Anthropic and Google for weeks about joint safety work. Chris Lehane, OpenAI’s global affairs chief, argued that voluntary industry standards should supplement rather than displace “mandatory federal safeguards and democratic oversight.” The company has also backed outside verification of top labs’ safety practices.1
The talks expose an awkward tension: the companies are fierce commercial rivals, yet all face pressure to show they can govern increasingly capable models. Microsoft chief Satya Nadella framed the task as both social and technical, calling for work to spread AI’s benefits, earn communities’ permission and ensure “AI safety and control.”
2
But the emerging consensus is far from universal. Reports say some participants have raised antitrust concerns if coordination suppresses competition; Amodei has floated a narrow government waiver, while Lehane says one is unnecessary. President Donald Trump and adviser David Sacks, meanwhile, have dismissed existential-risk arguments and warned a U.S. slowdown could advantage China.3
The sharpest industry critique comes from Meta AI scientist Yann LeCun, who amplified the argument that Anthropic and OpenAI may use safety regulation to “outlaw competitive open source models.”
4 The safety coalition now has to prove that cooperation protects the public without quietly protecting incumbents.