SAN FRANCISCO — OpenAI, Anthropic and Google DeepMind on Thursday formally announced a joint AI standards body to develop shared benchmarks for model safety, evaluation and interoperability.

The move follows reporting earlier this week that the firms were working together on such an organisation. Executives at all three companies had signalled growing pressure to demonstrate self-regulation as the US Department of Justice weighs its own investigation into the sector. Attorney General Todd Blanche said on Tuesday the DOJ would examine competition and safety practices among leading AI developers.

The new body will publish common testing protocols for frontier models, including evaluations for cybersecurity misuse, biological risk and deceptive behaviour. Microsoft and Amazon, which hold major stakes in the sector through partnerships and cloud infrastructure, were named as founding supporters.

Participation will be open to other developers meeting membership criteria.

The announcement drew immediate scrutiny from critics who argued voluntary industry frameworks cannot substitute for binding government oversight. Consumer and civil-society groups, including the Center for AI Safety, said standards written by the largest firms risked entrenching their market positions rather than protecting the public.

The launch positions the companies ahead of anticipated federal action and mirrors earlier self-regulatory efforts such as the Frontier Model Forum. Whether regulators in Washington and Brussels accept the framework as adequate will shape the next phase of AI governance.