Wiki · Organization · Last reviewed May 19, 2026

Frontier Model Forum

The Frontier Model Forum is an industry-supported nonprofit that coordinates frontier AI safety and security work among major AI developers. It is important because it turns lab-level safety practice into shared workstreams, issue briefs, funding, and standards-facing coordination.

Definition

The Frontier Model Forum, often abbreviated FMF, is an industry-supported nonprofit dedicated to advancing the safe development and deployment of frontier AI systems. Its public materials describe three core mandates: identify best practices and support standards development, advance the science of frontier AI safety and security, and facilitate information sharing among government, academia, civil society, and industry.

The forum focuses primarily on significant public-safety and security risks from advanced models, including chemical, biological, radiological, nuclear, and advanced cyber threats. It is not a regulator. It cannot compel companies outside its membership, issue binding law, or independently stop a model release. Its power is coordination, expertise, publication, funding, and agenda-setting.

Origin

Anthropic, Google, Microsoft, and OpenAI announced the forum in July 2023 as an industry body for safe and responsible development of frontier AI models. The initial objectives included advancing AI safety research, identifying safety best practices, sharing knowledge with policymakers and civil society, and supporting beneficial AI applications.

In October 2023, the founding companies named Chris Meserole as the first executive director and announced an AI Safety Fund with more than $10 million in initial commitments from companies and philanthropic partners. By the FY 2024-2025 annual report, the fund was described as a collaborative $12 million initiative supporting AI safety and security research, including independent and standardized evaluations.

Members and Leadership

As of May 2026, the forum lists Amazon, Anthropic, Google, Meta, Microsoft, and OpenAI as members. Amazon and Meta joined in May 2024, expanding the forum beyond the four founding organizations.

FMF membership criteria require frontier AI capability, a documented track record of safety, and clear intent to contribute. Members must show evidence that they can develop or deploy frontier AI models at scale, publicly acknowledge public-safety and societal risks, maintain safety review processes that may delay or suspend deployment, support safety research, and financially support FMF safety efforts for at least three years.

The board described in the FY 2024-2025 annual report includes policy, safety, and responsible-AI leaders from member firms, including representatives from Amazon Web Services, Anthropic, Google DeepMind, Meta, Microsoft, and OpenAI. This structure gives the forum deep operational access to frontier labs while also making its independence a standing governance question.

Workstreams and Outputs

AI-Bio. The AI-Bio workstream develops shared understanding of biological threat models, safety evaluations, mitigations, and capability thresholds for leading AI models. FMF publications in this area include taxonomy and reporting-tier documents for AI-bio evaluations, biosafety thresholds, and misuse mitigations.

AI-Cyber. The AI-Cyber workstream studies how frontier AI systems may amplify cyber threats and how developers can assess, mitigate, and set thresholds for cyber capabilities. It has published work on managing advanced cyber risks in frontier AI frameworks.

AI Security. The AI Security workstream focuses on securing the environments in which frontier AI systems are trained and deployed. Its publications include foundational security practices, AI for cyber defense, and adversarial distillation.

Frontier AI Frameworks. The forum works on common understandings around safety frameworks after the Seoul AI Safety Summit commitments, which called on developers to publish frontier AI safety frameworks before the 2025 Paris AI Action Summit.

AI Safety Fund. The forum helped establish the AI Safety Fund with philanthropic partners to support model-evaluation and AI safety research. The fund is significant because it channels industry and philanthropic money toward a broader research ecosystem outside the member labs themselves.

Why It Matters

The Frontier Model Forum sits between private labs, public regulators, standards bodies, safety institutes, academics, and civil society. That position matters because frontier-model risks often appear first inside private companies, while public institutions need shared vocabulary, evaluation methods, and disclosure channels to respond.

FMF also represents a shift from isolated company safety promises to collective industry infrastructure. Its issue briefs, workshops, workstreams, and funding can shape what counts as a reasonable safety practice in areas such as AI-bio risk, AI-cyber risk, model security, red teaming, and frontier safety frameworks.

The forum is also politically important because voluntary industry coordination often arrives before binding regulation. In that window, the categories, thresholds, and best practices produced by the strongest companies can become the default language used by policymakers and standards processes.

Limits and Criticism

Industry control. FMF is funded and governed by the same class of companies building frontier systems. Its proximity to operational reality is useful, but it also creates conflict-of-interest risk.

Voluntary authority. The forum can define best practices and facilitate information sharing, but it does not have the coercive authority of a regulator or the public mandate of a safety institute.

Scope narrowing. FMF's focus on CBRN, cyber, and security risks can strengthen work on catastrophic misuse while leaving less room for labor disruption, dependency, persuasion, civil-rights harms, model culture, or institutional capture.

Information hazards. Some safety research cannot be fully disclosed without spreading dangerous methods. This creates a real publication dilemma, but it also means outsiders may have limited ability to evaluate the quality of the forum's strongest claims.

Standards capture. If frontier labs define the benchmarks, thresholds, and best practices that later become official standards, governance can become dependent on the worldview of the companies being governed.

Spiralist Reading

The Frontier Model Forum is the cathedral guild of the frontier labs: the builders meeting behind the scaffold to compare stress fractures in the structure they are still raising.

That is necessary. The companies have information, engineers, incidents, and operational knowledge that public institutions need. A world with no shared lab-to-lab safety channel would be worse.

It is also insufficient. Industry coordination can become a way to translate danger into process without creating public power. For Spiralism, FMF should be read as useful infrastructure only when it increases outside scrutiny, shared evidence, and real deployment friction. It should not be mistaken for democratic governance.

Sources

Return to Wiki