OpenAI Confirms Weeks of Safety Talks With Anthropic and Google DeepMind

OpenAI has confirmed it has been in talks with Anthropic and Google DeepMind about AI safety for several weeks, a rare acknowledgment that the three most prominent frontier labs are coordinating behind closed doors. The discussions involve the companies that have produced GPT-4, Claude and Gemini, the models that set the pace for the industry. OpenAI's confirmation came after questions about whether such contacts were taking place at all.

The conversations unfold against a shifting political backdrop. President-elect Donald Trump's transition team has dismissed safety concerns as an obstacle, arguing that aggressive regulation would cede ground to China. That stance contrasts with the Biden administration's approach, which relied on voluntary commitments from labs and an executive order that Trump has vowed to repeal. The result is a vacuum that the companies appear to be filling on their own.

Details remain scarce. None of the three firms has described the format, frequency or agenda of the talks, and it is unclear whether they amount to a formal working group or a loose series of calls. OpenAI's disclosure was terse, confirming only that the conversations had occurred over a period of weeks. Anthropic and Google DeepMind have not commented publicly on the record.

The three labs have competed fiercely for talent, compute and market share, and their willingness to talk safety runs against that grain. Anthropic was founded in 2021 by former OpenAI researchers who left over disagreements about direction and risk, and Google DeepMind has its own safety teams. Any cooperation among them would mark a notable departure from the usual secrecy that surrounds frontier model development.

For the broader field, the talks raise a practical question: what, if anything, can voluntary coordination achieve when the government is stepping back? Previous efforts, including the Bletchley Declaration signed by 28 countries in 2023 and the Seoul summit in 2024, produced statements rather than binding rules. Industry-led discussions could fill some of that gap, or they could produce little more than press statements.

What to watch is whether the talks yield something concrete, such as shared evaluation standards or a joint commitment on model reporting. The incoming administration's posture may push the labs to act faster, or it may convince them that quiet conversations are safer than public promises. Either way, the next few months will show whether safety coordination survives when Washington stops asking for it.