AI leaders explore joint safety standards body

Anthropic, OpenAI and Google have been holding private working-group talks since July to explore forming a joint, industry-led body focused on safety standards for advanced AI systems. The initiative centers on frontier AI models, where companies are seeking ways to coordinate on testing and risk management for increasingly capable technologies.

These discussions aim to establish common ground on how powerful AI models should be evaluated before release, and how risks should be assessed across different organizations. The talks remain confidential in detail, but they mark a notable step toward more structured collaboration among leading AI developers.

The current conversations build on a proposal made on July 14 by Google DeepMind CEO Demis Hassabis for a U.S.-based Frontier AI Standards Body. That concept drew on the model of financial industry self-regulatory organizations such as FINRA, suggesting an independent entity dedicated to overseeing standards for high-risk AI systems.

This earlier proposal is cited as groundwork for the present working group, helping frame what an industry-led standards body might look like. It provides a template for potential governance structures without yet defining the concrete rules such a body would enforce.

Reporting indicates that the talks among Anthropic, OpenAI and Google are still at an exploratory stage. Participants have not announced a formal organization, agreed charter or finalized rulebook, and no enforcement mechanisms have been described publicly.

No common standards, pre-release testing protocols or shared auditing frameworks have been published so far. The working group is focused on whether, and how, a coordinated approach could be developed, rather than implementing any immediate regulatory scheme.

The emerging collaboration among the three companies sits within a larger debate over how AI should be governed. The idea of an industry-led standards body highlights interest in structured self-regulation, even as questions remain about the future role of national regulators and legislatures.

These talks underscore that leading AI developers are examining options for joint safety practices while navigating uncertainty about future regulation. Their outcome could influence how frontier AI risk management is organized, though no concrete institutional framework has yet been put in place.