Anthropic Partners With METR and SecureBio to Evaluate AI Safety Risks
Anthropic has enlisted nonprofits METR and SecureBio to assess its AI risk levels and safety practices, with more partners potentially on the way.
Anthropic has brought on two nonprofit organizations — METR and SecureBio — to independently evaluate the risks and safety practices surrounding its artificial intelligence systems, the company announced. The move signals a growing push within the AI industry toward external accountability as models become increasingly powerful and their potential harms draw heightened public and regulatory scrutiny.
METR, which focuses on evaluating the capabilities and risks of advanced AI models, and SecureBio, which examines biosecurity threats posed by emerging technologies, each bring specialized expertise to the partnership. By enlisting organizations with distinct but complementary threat domains, Anthropic is broadening the scope of its safety assessments beyond what an internal review process alone could capture.
Read more How Much Apple Stock Does New CEO John Ternus Own? →
The arrangement reflects a wider trend among leading AI developers to demonstrate transparency through third-party audits, a practice that regulators in both the United States and Europe have pushed for with increasing urgency. Independent evaluations can help identify capability thresholds or misuse vectors that developers may be too close to their own work to spot objectively.
Anthropolic is also in early discussions with additional organizations about similar roles, including open-source AI platform Hugging Face, suggesting the company intends to expand its external safety network further. While the scope and timeline of those conversations remain unclear, the outreach to Hugging Face — a major hub for AI model sharing and community research — could eventually extend safety scrutiny to a broader ecosystem of users and applications.
The selection of nonprofits rather than commercial auditors may reflect a deliberate choice to prioritize mission-driven credibility over purely transactional reviews, though the effectiveness of any such program will ultimately depend on the independence and access granted to evaluators. Continue reading at US Top News and Analysis
