OpenAI, Anthropic, and Google DeepMind have been holding private talks about AI safety for several weeks, according to Chris Lehane, OpenAI’s global policy chief. He told reporters on Tuesday while in Washington to work with U.S. lawmakers on catastrophic AI risks. The disclosure confirms earlier reporting and follows Anthropic CEO Dario Amodei’s essay published Saturday, which urged the industry to cooperate in slowing frontier AI development to avoid catastrophic outcomes. OpenAI’s Sam Altman, Google’s Demis Hassabis, and Elon Musk all publicly supported Amodei’s call; Altman said OpenAI would join Anthropic in embedding third party evaluators into the company to monitor safety.

Lehane’s remarks put a formal stamp on coordination that had been hinted at for weeks. In an interview with Fortune late last week, Altman suggested he had been in private discussions with other AI leaders. The Information then reported that the three firms were working together to create an industry standards body, something Altman reportedly told staff would need to happen without U.S. government support. Hassabis had previously called on the U.S. in July to establish a standards body with the power to screen advanced models and coordinate industry-wide slowdowns if dangers escalated.

The coordination raises legal questions. Some, including Altman, have noted the risk of antitrust violations if the collaboration is found to suppress competition. Amodei’s essay proposed a narrow government waiver for safety coordination, but Lehane reportedly said the companies do not need one. At the same Tuesday meeting, Lehane said OpenAI supports a provision in the FRONTIER Act that would require top frontier labs to allow independent verification organizations inside their companies to ensure safe model development.

The political context remains divided. President Trump has dismissed safety concerns as a hoax and argued that slowing down would help China, while his AI advisor David Sacks called existential-risk fears overblown. TechCrunch reached out to Anthropic, Google, and OpenAI for comment.