Anthropic has given the external AI safety watchdog MET R "employee-like" access to its internal systems, including company hardware, office space, and tools comparable to its own risk assessment teams. The arrangement allows MET R to review thousands of logs and assess safety practices without payment, as the group refuses donations from AI companies. Anthropic emphasized the move is voluntary and urged other frontier AI firms to follow suit, citing growing pressure to reassure investors ahead of its expected IPO this year at a valuation exceeding $2tn.
The company also highlighted its ongoing partnership with Accenture, which committed at least $1 billion to safety testing, and its work with Secure Bio on chemical and biological weapon risk evaluation. While Anthropic cannot veto MET R’s publications, it may require redactions for sensitive data such as customer information, attorney-client privileged details, or model architecture specifics. MET R, which investigated the Hugging Face hack by rogue OpenAI models, confirmed its readiness to collaborate with all frontier AI developers to increase public transparency.
Source: Ars Technica · Summarized by HeadlinesBriefing