OpenAI in talks with Google and Anthropic over shared AI safety standards

OpenAI has confirmed it is in active discussions with its chief rivals, Anthropic and Google, to find ways to collaborate on artificial intelligence safety and coordinate how quickly they release their most advanced models.

The talks represent a rare moment of unity among the three tech giants, who are otherwise locked in an expensive, high-stakes race to dominate the generative AI market. An OpenAI spokesperson confirmed the discussions, which have been ongoing since the summer.

Why are the rivals talking?

The sudden willingness to cooperate follows growing anxiety within the industry about the sheer speed at which AI models are advancing. In July, Demis Hassabis, the chair of Google DeepMind, published an essay suggesting the creation of a US-led self-regulatory body with federal oversight, similar to the financial industry's regulatory framework.

The push gained momentum when Anthropic Chief Executive Dario Amodei published his own essay urging AI developers to slow down the release of their frontier systems.

Both Hassabis and OpenAI Chief Executive Sam Altman publicly endorsed Amodei's warning. Altman noted on social media that the concept of pacing model releases has become a "primary topic" of internal discussion at OpenAI, hinting that the company would share its plans in the near future.

What would the safety standards look like?

While the details of any agreement are still being hammered out, the discussions focus on creating voluntary, industry-led standards that can be deployed quickly.

Chris Lehane, OpenAI's chief global affairs officer, recently argued that the current leap in AI capabilities requires immediate, voluntary action from developers. Writing in a blog post, Lehane said the industry should not wait for slow-moving government legislation to take effect. He clarified that any voluntary framework would complement, rather than replace, official federal safeguards and democratic oversight.

The move toward self-regulation comes as Google DeepMind's own research, compiled with MIT FutureTech, highlights the accelerating pace of AI adoption. Their data shows that nearly half of scientists surveyed in the US and UK now use AI daily, saving an average of seven hours a week. While this has boosted productivity, it has also created a bottleneck of unverified hypotheses, highlighting the unpredictable ways advanced technology can outpace human systems.

Tackling biological weapons in Congress

Alongside these private industry negotiations, OpenAI is also stepping up its public lobbying efforts to address worst-case scenarios. The company has thrown its support behind bipartisan legislation in the US Congress designed to prevent AI models from being used to develop biological weapons.

The threat of AI-assisted bioweapons has become a central concern for Washington lawmakers, who worry that highly capable models could soon lower the technical barrier required to engineer deadly pathogens. By backing these federal bills, OpenAI is seeking to position itself as a responsible partner in national security, even as it negotiates voluntary safety limits with its biggest commercial competitors.

Neither Google nor Anthropic have commented publicly on the progress of the safety talks.