Frontier AI Labs Discuss Shared Body to Oversee Safety Risks

OpenAI is working with Anthropic and Google DeepMind to establish a shared organization focused on artificial-intelligence safety risks, turning a recent wave of warnings from frontier-lab leaders into a tentative coordination effort.

OpenAI global policy chief Chris Lehane disclosed the discussions to reporters on September 15, according to TechCrunch and Bloomberg reporting. A separate account carried by Qatar News Agency described the initiative as a body intended to oversee AI risks. The reports did not identify a finalized charter, launch date, funding model, or enforcement mechanism.

What changed

The development is significant because the leading labs have generally competed over models, talent, infrastructure, and market share while relying on separate internal safety programs. A joint body would create at least a channel for sharing evaluations, coordinating standards, and comparing safeguards before highly capable systems are released.

The discussions follow Anthropic CEO Dario Amodei’s September 12 call for the industry to slow the development of frontier models long enough for safety measures to catch up. OpenAI CEO Sam Altman, Google DeepMind CEO Demis Hassabis, and SpaceX AI chief Elon Musk have expressed support for some form of coordination, although their positions are not identical.

Lehane also said OpenAI supports legislation that would require major frontier developers to allow independent verification organizations to assess whether their systems are being developed safely. That proposal points toward a model in which companies retain control of development but permit outside testing and review.

Why it matters

A shared institution could help address a basic weakness in voluntary AI governance: each company is asked to judge its own systems while competing to deploy them quickly. Common testing protocols or third-party audits could make safety claims more comparable and expose failures that a single lab might miss.

It could also give policymakers a technical counterpart. Governments are being asked to regulate systems whose capabilities change faster than formal standards can be written. A standing organization involving the largest developers might provide common terminology, incident reporting, and benchmark results that regulators could use.

But the proposal arrives amid a sharp political split. President Donald Trump has dismissed calls for a slowdown, warning that restrictions could give China an advantage. Meta CEO Mark Zuckerberg has likewise argued that individual labs have sufficient incentives and liability exposure to develop safely on their own. Those objections could limit participation or push the proposed body toward nonbinding guidance.

What remains uncertain

The central question is whether the group would be a standards forum, an audit network, an incident-reporting clearinghouse, or something with authority to delay deployments. The companies have not publicly explained how disagreements would be resolved, whether safety findings would be disclosed, or whether smaller labs and civil-society groups would participate.

The talks therefore mark a meaningful change in posture, not yet a new regulatory regime. Their credibility will depend on whether the labs accept independent scrutiny and publish measurable commitments rather than simply coordinating public messaging.