OpenAI, Google, Microsoft and Anthropic Form Frontier Model Forum

OpenAI, Google, Microsoft and Anthropic announced today the creation of the Frontier Model Forum, an industry body dedicated to promoting the safe and responsible development of so-called "frontier models": the largest and most capable AI systems in existence, those that exceed the capabilities of any prior model and can perform a wide range of complex tasks.

It's the first time these four direct competitors — who battle each other client by client in the generative AI market — have formally organized to coordinate common safety standards, rather than each setting its own.

What the Forum Is—and Isn't

The Frontier Model Forum is not a regulator and has no power to impose anything on its members or on third parties. It's a self-regulation initiative: the labs themselves define its goals and decide which practices to share. According to the joint announcement from the four companies, the forum has set four concrete tasks: advancing AI safety research to minimize risks, identifying and sharing best practices for frontier model development, collaborating with policymakers, academics and civil society by sharing knowledge on trust and safety risks, and supporting the use of AI to address major societal challenges, from climate change to early disease detection.

Membership remains open to other organizations, provided they meet three criteria: they must develop and deploy frontier models, demonstrate a strong commitment to the safety of those systems, and be willing to actively contribute to the forum's work. In practice, this leaves out — for now — smaller labs or companies that don't operate at the frontier of capability, though the forum's design itself invites its membership roster to grow.

The Backdrop: A Week of Self-Regulation Gestures

Today's announcement comes just five days after the White House unveiled, on July 21, voluntary safety commitments adopted by seven major AI companies — including the Forum's four founders, alongside Amazon, Meta and Inflection AI. Those commitments included subjecting models to external testing before release, investing in cybersecurity to protect model weights, and watermarking AI-generated content.

The Frontier Model Forum can be read as the institutional follow-through on that same push: turning one-off promises made to the U.S. administration into a permanent structure meant to outlast a single week's news cycle.

Why Now, and Why These Four

The launch of ChatGPT in late 2022 and the release of GPT-4 in March of this year fueled both commercial interest in generative AI and public concern over its risks: disinformation, malicious use, or future capabilities that even the labs themselves can't predict with certainty. Against that backdrop, governments in the United States, the European Union and the United Kingdom have accelerated their own regulatory processes, and the major labs have a clear incentive to appear proactive: arriving at the regulatory table with their own rules already sketched out tends to beat having rules imposed from outside.

That it's specifically OpenAI, Google, Microsoft and Anthropic leading this initiative is no accident: along with Meta, they are the only labs currently training and deploying the largest, most expensive models on the market. Meta, however, is not among the founders, despite having signed the White House commitments just last week.

What Remains to Be Seen

Today's announcement sets out goals and intent, not mechanisms. Neither the forum's working timeline nor how compliance with practices agreed upon by its members will be measured has been detailed yet. Nor is there, for now, any formal participation by governments or independent bodies in its governance: the labs are evaluating themselves.

That's the underlying tension of any self-regulatory forum: its usefulness hinges on whether the companies within it are willing to constrain themselves in a market where the speed of shipping new models is, for now, the chief competitive advantage.