Summary
On July 26, 2023, Anthropic, Google, Microsoft and OpenAI announced the Frontier Model Forum, an industry body for the "safe and responsible development of frontier AI models." The forum said it would advance AI safety research, identify safety best practices, share information with policymakers and other companies, and support uses of AI for problems such as climate change and cancer. It was announced five days after the four companies joined the White House voluntary commitments, and a researcher quoted by The Guardian warned that oversight of AI must not fall to regulatory capture.
What Happened
The four companies published a joint announcement on their own sites. They defined frontier models as "large-scale machine-learning models that exceed the capabilities currently present in the most advanced existing models, and can perform a wide variety of tasks." Membership was open to organizations that developed and deployed frontier models, demonstrated "strong commitment to frontier model safety" and were willing to contribute to the forum's joint work.
For its first year the forum named three areas of work. It would identify best practices and safety standards, coordinate research in areas such as adversarial robustness, mechanistic interpretability, scalable oversight and anomaly detection, with an initial focus on a public library of technical evaluations and benchmarks, and set up "trusted, secure mechanisms" for sharing information about AI safety and risks among companies and governments. The founders said they would establish an advisory board, a charter, governance and funding, and would consult civil society and governments on the forum's design. The announcement said the forum would support existing efforts such as the G7 Hiroshima process, the OECD's work on AI risks and the US-EU Trade and Technology Council.
Microsoft president Brad Smith said that "companies creating AI technology have a responsibility to ensure that it is safe, secure, and remains under human control." Anthropic chief executive Dario Amodei said the forum would "play a vital role in coordinating best practices and sharing research on frontier AI safety." The Guardian quoted Andrew Rogoyski of the University of Surrey, who warned against "regulatory capture" and said, "I have grave concerns that governments have ceded leadership in AI to the private sector, probably irrecoverably."
Why It Matters
The forum gave institutional form to the idea, central to the White House commitments, that frontier developers could coordinate on safety practices themselves. Its founding membership was limited to companies building the largest models, and its definition of frontier models was written by those companies, which made it both the most direct channel between the leading labs on safety questions and open to the charge of regulatory capture.
The forum's definition arrived as the term was entering policy language. The UK's Bletchley summit in November 2023 was organized around "frontier AI," and California's SB 53 later set obligations for frontier developers, using the same category of the most capable general-purpose models.
§ How to read the metadata
- Landmark
- Fundamentally alters the trajectory; 2–5 per year.
- Major
- Meaningfully shifts the landscape; 2–4 per month.
- Notable
- Worth documenting; significance can be upgraded later.
- Confidence
- High = primary sources corroborate. Medium = credible secondary only. Low = provisional. Disputed = credible sources disagree.
- Contestation
- Uncontested = no formal challenge. Contested = at least one challenge filed and not resolved by a correction. Corrected = a challenge was upheld and the entry corrected; see its revision history. Superseded = replaced by a later entry. Unresolved = dispute still open.