Anthropic embeds METR evaluators and creates public benefit fund

Anthropic has announced a dual initiative to strengthen its governance and societal impact, granting the nonprofit METR "employee-like access" to its internal systems while committing 5% of its shares to a new public benefit fund. According to Ars Technica, this marks the first concrete step in implementing pledges made by industry leaders to embed external auditors in leading AI labs. The move comes as the company prepares for an initial public offering expected this year at a valuation exceeding $2 trillion, with CEO Dario Amodei describing the current industry climate as a "fundamentally a crisis of trust." TechCrunch reports that the public benefit fund is structured as a wholly-owned entity, designed to provide "increased flexibility and scale" for projects benefiting society.

The details

The arrangement with METR involves onboarding six of the organization's employees into Anthropic’s office, providing them with company hardware, office space, and access to internal tools and systems that are "roughly on par with what our own risk assessment teams have access to." Anthropic explicitly stated that it is not paying METR for this access; instead, METR is funded by individual donations, philanthropic organizations, and government bodies such as the UK’s AI Security Institute. This structure is intended to avoid conflicts of interest, drawing a parallel to bank supervision where regulators assign resident examiners to large financial institutions. Anthropic noted, "If a bank’s risk-taking could threaten the public, third parties try to see that in advance." While Anthropic can require redactions for customer data, attorney-client privilege, or model architecture details, it cannot veto the publication of METR’s analysis.

Simultaneously, Anthropic is establishing the public benefit fund with an initial commitment of 5% of its shares. The company has outlined an option to contribute an additional 2% for every 50% increase in Anthropic’s value. Potential uses for the fund include "large donations to traditional nonprofits or direct distributions to specific populations." Anthropic also highlighted its existing $200 million commitment to studying the impact of AI on the job market, suggesting that the fund will allow for "bolder approaches" given the pace of AI progress. The company is also continuing its partnership with Accenture, with each entity committing at least $1 billion to new safeguards, and working with SecureBio to review chemical and biological weapon risks.

Context

This announcement follows a rare moment of unity in September when Dario Amodei, Sam Altman, and Elon Musk rallied behind Amodei’s proposal to embed external auditors in leading labs. While OpenAI has yet to lay out specific plans for evaluators, Anthropic is moving first to implement these pledges. Critics have argued that such arrangements could lead to conflicts of interest similar to issues in the accounting sector, advocating instead for greater government regulation. However, Anthropic argues that public disclosure of safety work could benefit malicious actors or competitors, necessitating a controlled, third-party evaluation model. METR, which previously investigated the hack of Hugging Face by rogue OpenAI models, has stated it is "ready to work with all frontier AI developers to bring better evidence to the public."

What it means for developers

For developers and enterprise users, this shift signals a more rigorous internal risk assessment process that may influence model capabilities and safety guardrails. The involvement of METR, a group with experience in investigating model misbehavior, suggests that future model releases may undergo more stringent third-party scrutiny. While this does not immediately change API endpoints or pricing, it indicates a strategic pivot toward transparency and accountability that could affect how Anthropic communicates safety incidents and model limitations. Developers should watch for how these external evaluations are published, as they may provide independent benchmarks for model safety that go beyond Anthropic’s internal claims.

What to watch

• Governance Structure: Anthropic has promised to share more information about the structure and governance of the public benefit fund in the coming months.
• METR Reports: The first published analysis from the embedded METR team will be a key indicator of the depth and independence of the evaluation process.
• IPO Timeline: The success of the initial public offering, expected at a valuation over $2 trillion, will test whether investors accept safety-focused governance as a value proposition.
• Competitor Response: Whether OpenAI and other frontier labs follow suit with similar third-party evaluator arrangements will determine if this becomes an industry standard.