OpenAI, Anthropic and Google DeepMind are coordinating on AI safety

OpenAI’s global policy chief Chris Lehane said the company has been working with Anthropic and Google DeepMind on AI safety for several weeks. Reuters relayed the disclosure from Bloomberg, while TechCrunch separately reported Lehane’s remarks from a Washington briefing.

The reporting describes coordination around safety concerns rather than a merger of research programmes. Lehane said OpenAI did not believe the current discussions required a special antitrust waiver, suggesting the companies view the work as limited enough to proceed within ordinary competition rules.

The talks are emerging during a period of unusually public concern from frontier-lab leaders about catastrophic-risk controls, model evaluations and security. Anthropic chief executive Dario Amodei recently argued that the industry should slow the frontier enough for safeguards and independent assessment to keep pace.

That makes the coordination notable even before its practical output is clear. OpenAI, Anthropic and Google DeepMind compete directly on models, enterprise customers and developer platforms, so sustained work across organisational boundaries would represent a stronger form of cooperation than each company publishing its own safety framework.

There is no public joint charter, technical standard, shared evaluation protocol or governance structure attached to the discussions in the reports available so far. Reuters also noted that it had not independently verified Bloomberg’s original account and that the companies had not responded to its requests for comment at publication time.

The useful signal will be whether the talks produce something inspectable: common evaluation criteria, incident-sharing rules, third-party testing, deployment thresholds or a standing cross-lab body. Until then, the status should remain Corroborated rather than Confirmed because the evidence comes from matching reporting around Lehane’s remarks, not a joint first-party announcement from the three labs.