The largest developers of artificial intelligence have begun coordinating efforts on the safety of their models amid growing concerns about the risks posed by the development of advanced systems. OpenAI, Anthropic, and Google DeepMind have already been in talks for several weeks, while at the same time the debate in the United States is intensifying over whether AI development should be slowed down at all.

Chris Lehane, OpenAI’s head of global policy, spoke about the cooperation on Tuesday. According to TechCrunch, citing Bloomberg, he is currently in Washington working with U.S. lawmakers on issues related to the potentially catastrophic risks of AI.

Competitors have decided to discuss safety together

It became known that the largest AI companies might coordinate their actions after remarks by Anthropic CEO Dario Amodei. In an essay published on September 12, he called on developers of advanced models to jointly slow the pace of the race toward ever more powerful AI in order to prevent catastrophic consequences.

The idea was supported by executives from several of the industry’s biggest companies. OpenAI’s Sam Altman said his company was ready to follow Anthropic’s proposal and bring in third-party experts who could independently assess model safety. Google DeepMind’s Demis Hassabis also backed the initiative. Elon Musk voiced his support as well.

It has now become clear that this was not just about public statements. OpenAI, Anthropic, and Google DeepMind really have been discussing safety issues among themselves for several weeks.

At the same time, such cooperation creates an unexpected legal problem for competitors. Companies that are simultaneously competing for leadership in the AI market cannot freely agree on joint actions without limitations: such coordination could potentially raise antitrust concerns.

Amodei therefore proposed creating a special narrow exception in legislation that would allow AI companies to cooperate specifically on safety issues. According to Lehane, OpenAI believes a separate government carve-out for such talks is not necessary.

Companies want to create common rules

Recent developments show that the discussion may go even further. It was previously reported that OpenAI, Anthropic, and Google are working on creating an industry body that would deal with AI safety standards.

The idea is for the industry to have common rules for evaluating the most powerful models. Altman reportedly told OpenAI employees that such a mechanism should emerge even without the participation of the U.S. government.

Back in July, Hassabis called on Washington to create a separate body that would serve as an overseer of AI development. Its possible powers would include inspecting the most advanced models and coordinating a slowdown in their development if the level of danger begins to rise.

Thus, this is no longer simply an exchange of views between competitors. The companies are trying to develop a mechanism that would allow the industry to establish uniform safety criteria for systems whose capabilities are rapidly approaching limits that even the developers themselves cannot yet always reliably assess.

In Washington, they believe slowing down is not an option

At the same time, this idea has a serious political opponent. Donald Trump’s administration opposes tighter AI regulation and warnings about possible catastrophic risks.

Trump himself has called concerns about AI safety a hoax and said that slowing the industry’s development could give China an advantage in the technological race. His AI adviser David Sacks has also said that concerns about the existential risks of artificial intelligence are exaggerated.

This creates an unusual situation: the largest AI companies, competing for leadership in one of the fastest-growing technology sectors, are simultaneously trying to agree on rules that could limit the speed of this race. The U.S. authorities, by contrast, fear that such restrictions would weaken America’s position in its rivalry with China.

OpenAI backed independent inspections

At the same meeting on Tuesday, Lehane said that OpenAI supports a provision of the FRONTIER Act that would require the largest labs developing advanced AI models to allow independent organizations to inspect the safety of their systems.

This is an important difference from a model in which the developers themselves determine how safe the AI they created is. Independent review should give outside specialists the opportunity to assess the development process and identify potentially dangerous problems.

It is still unknown whether the current talks between OpenAI, Anthropic, and Google will lead to formal industry rules. But the very fact of their cooperation shows how seriously the largest developers have begun to take the problem: competition for the most powerful models continues, yet companies are increasingly concluding that some safety issues cannot be effectively addressed alone.