Chris Lehane, OpenAI’s head of global policy, confirmed to reporters on September 15 that OpenAI, Anthropic, and Google DeepMind have been holding joint discussions on AI safety for several weeks. This marks the first time the industry’s three largest rivals have publicly acknowledged such collaboration.
The revelation comes just days after Anthropic CEO Dario Amodei published an essay calling on the entire industry to collectively slow down the development of frontier AI to avert catastrophic risks. The essay quickly drew support from Sam Altman (OpenAI), Demis Hassabis (Google DeepMind), and Elon Musk (SpaceXAI). Altman said OpenAI would join Anthropic in allowing third-party evaluation organizations to monitor internal safety.
Formation of an industry standards body
According to The Information, the three companies are working together to establish a standards body for the AI industry. Altman told employees that this would have to happen without the support of the U.S. government. Hassabis previously called on the U.S. to create a new AI oversight agency with the authority to review the most advanced models and coordinate industry-wide slowdowns if dangers are detected.
This is an unprecedented move. Historically, the largest AI companies have fiercely competed and rarely collaborated publicly on safety. The shift reflects genuine concern within the industry, especially after an Anthropic researcher resigned and warned about “gambling with lives” over self-improving AI.
OpenAI also stated its support for a provision in the draft FRONTIER Act requiring frontier labs to allow “independent verification organizations” inside the company to ensure models are developed safely.
Antitrust questions
Coordination among the industry’s three largest companies raises serious legal questions. Some, including Altman, have noted that talks of this nature could run afoul of antitrust laws if viewed as agreements to restrain competition. Amodei’s essay proposed a narrow government exemption to permit safety coordination, but Lehane said the companies do not need such an exemption.
It is a dilemma: slowing down together could bolster safety, but it could also be exploited to entrench dominant positions and hinder smaller rivals. The line between “safety cooperation” and an “industrial cartel” is very thin.
Washington says no to safety
While the companies are sitting down together on their own, the Trump administration is heading in the exact opposite direction. The president has called AI safety concerns a “hoax” and argued that any slowdown would allow China to pull ahead in the race. AI adviser David Sacks, a longtime investor with personal investments in several industry companies, said existential risk from AI is “overblown.”
This contrast creates a picture with 2 opposing extremes, where AI companies worry more about safety than the government that should be overseeing them. Can the industry establish and abide by safety standards without external enforcement mechanisms? The answer could shape the entire trajectory of AI development over the coming decade. And if the U.S. government fails to act, the companies will write the rules for themselves—a situation that even insiders acknowledge is problematic.