OpenAI, Anthropic and Google Move AI Safety Talks Into the Open

Introduction

Cooperation among frontier AI companies on safety is moving from private conversations into public view. Chris Lehane, OpenAI’s global policy chief, told reporters that OpenAI has been working with Anthropic and Google DeepMind on AI safety for several weeks. The disclosure comes as senior U.S. officials and advisers dismiss or downplay catastrophic AI concerns, arguing that tighter safeguards could slow America’s ability to compete with China.

Key points

• Three major labs acknowledge ongoing contact. Lehane did not describe the discussions as a formal alliance or regulatory body. He did, however, confirm that the three companies have been exchanging views on AI safety, validating earlier reports of private conversations among industry leaders.
• Independent evaluation is emerging as a shared idea. Anthropic CEO Dario Amodei recently called on frontier AI companies to work together and slow the pace of development to reduce catastrophic risks. OpenAI CEO Sam Altman said OpenAI would join Anthropic in embedding third-party evaluators to monitor safety.
• A standards body may be under discussion. The Information reported that the three companies have been working toward an AI industry standards organization. Google DeepMind CEO Demis Hassabis previously called for a U.S.-backed watchdog able to screen advanced models and coordinate an industrywide slowdown if risks worsened. The material does not establish that such a body has been formally created.
• Coordination creates an antitrust dilemma. Altman and others have warned that cooperation among direct competitors could raise antitrust concerns if it affects competition. Amodei proposed a narrow government waiver for safety coordination, while Lehane reportedly said the companies do not need one.
• OpenAI backs independent verification. Lehane said OpenAI supports a provision in the FRONTIER Act that would require leading frontier labs to admit “independent verification organizations” to check whether models are being developed safely.

Why it matters

The significance of these talks is not that the companies have announced a binding slowdown. It is that safety is increasingly being treated as a shared infrastructure problem rather than an issue each laboratory can define alone. If every company sets its own evaluation rules, outsiders will struggle to compare the safety of different models or determine whether risk claims are credible. Yet extensive information-sharing among competitors could expose companies to competition-law scrutiny.

Independent evaluators and common standards could offer a middle path. Labs would still compete on capability and products, while allowing outside organizations to test models against a more consistent framework. The details will matter: who funds the evaluators, what access they receive, how confidential information is protected, and whether findings are disclosed publicly. Without meaningful independence and transparent procedures, a standards label could become little more than industry self-certification.

The political context makes the effort harder. President Trump has dismissed AI safety concerns and argued that slowing development could give China an advantage. His AI adviser David Sacks has likewise said existential-risk fears are overstated. Companies therefore face two opposing pressures: demonstrate that frontier development is responsible, while showing that safety reviews will not become a vehicle for suppressing innovation or competition. The next test will be whether private discussions can produce rules that are transparent, verifiable and legally defensible.

Source: TechCrunch AI