Frontier Model Forum Updates

OpenAI, Anthropic, Google, and Microsoft have appointed Chris Meserole as the first Executive Director of the Frontier Model Forum and launched a $10 million AI Safety Fund. These initiatives aim to ensure the safe and responsible development of frontier AI models through leadership, independent research funding, and the establishment of industry-wide technical standards.

Appointment of Executive Director

Chris Meserole, formerly the Director of the Artificial Intelligence and Emerging Technology Initiative at the Brookings Institution, has been named the first Executive Director of the Frontier Model Forum. In this role, Meserole is responsible for leading the Forum's mission to:

  • Advance AI safety research to minimize potential risks and promote responsible development.
  • Identify and establish safety best practices for frontier models.
  • Facilitate knowledge sharing with academics, policymakers, and civil society.
  • Support the use of AI to address major societal challenges.

The AI Safety Fund

To address the gap in academic research as AI capabilities accelerate, the Frontier Model Forum and several philanthropic partners have committed over $10 million in initial funding for a new AI Safety Fund.

Funding Sources and Administration

The fund is supported by Anthropic, Google, Microsoft, OpenAI, and philanthropic partners including the Patrick J. McGovern Foundation, the David and Lucile Packard Foundation, Eric Schmidt, and Jaan Tallinn. The fund will be administered by the Meridian Institute and overseen by an advisory committee of independent external experts, AI company representatives, and grantmaking specialists.

Objectives and Scope

The primary focus of the AI Safety Fund is to support independent researchers from academic institutions, research centers, and startups. The fund specifically targets the development of new model evaluations and red teaming techniques to test for potentially dangerous capabilities in frontier systems. This initiative fulfills part of the voluntary AI commitments signed at the White House earlier in 2023 to facilitate third-party discovery and reporting of vulnerabilities.

Technical Standards and Red Teaming

The Frontier Model Forum is working to establish a common baseline of definitions and processes to synchronize discussions between researchers, governments, and industry peers.

Standardizing Red Teaming

The Forum has released its first technical working group update, which provides a common definition of "red teaming" for AI. The Forum defines red teaming as "a structured process for probing AI systems and products for the identification of harmful capabilities, outputs, or infrastructural threats."

Responsible Disclosure Process

The Forum is developing a responsible disclosure process to allow frontier AI labs to share information regarding discovered vulnerabilities, dangerous capabilities, and associated mitigations. This process is informed by existing research conducted by Forum members on AI capabilities and mitigations within the realm of national security.

Future Roadmap

The Frontier Model Forum will establish an Advisory Board to guide its strategy and priorities. Upcoming milestones include the issuance of the first call for proposals for the AI Safety Fund and the release of further technical findings as they become available.

Sources