The Standards Authority for Frontier AI will establish security and safety benchmarks for advanced artificial intelligence models starting in late 2026 or early 2027. The independent group, formed by Google, OpenAI and Anthropic, will set qualifications for model auditors and create rules for reporting safety incidents. It will also support third-party testing of models before they are released and may perform its own safety and capability evaluations.

The three companies are launching the voluntary body to fill a regulatory vacuum after previous efforts to create a public-private governance framework stalled. This new organization will operate without government oversight, shifting the responsibility for establishing frontier AI standards to the private sector.