OpenAI, Anthropic, Google and Microsoft launch the Frontier Model Forum

OpenAI, Anthropic, Google and Microsoft announced the formation of the Frontier Model Forum, an industry body intended to advance safe development of the most capable AI models. The four companies described its purpose as identifying best practices for frontier-model safety, advancing safety research including in areas such as interpretability and adversarial robustness, and facilitating information-sharing among companies and with governments, academia and civil society.

The Forum arrived five days after the same core group, plus Amazon and Inflection, had made voluntary safety commitments at the White House. Where the White House commitments were a government-brokered pledge, the Forum was the industry’s own standing structure — a membership body with founding companies rather than an agreement extracted by officials, positioned to persist and set its own agenda over time rather than mark a single announcement.

Membership was initially restricted to companies developing models the founders judged to meet a “frontier” capability bar, an implicitly self-defined threshold that gave the founding labs discretion over who else could join. The Forum later added members including Amazon, Meta and additional labs, and it funded an AI Safety Fund intended to support external researchers.

As with other industry-led safety initiatives of the period, the Forum carried no regulatory authority and no mechanism to compel any member’s behaviour; its standards were the labs’ own, adopted and revised at their own discretion. Supporters characterised it as filling a gap ahead of slower government rulemaking; critics noted that a body founded and controlled by the companies it purported to oversee had an obvious structural conflict, and that its practical output over the following years — shared research findings and voluntary evaluation frameworks — remained non-binding throughout.