1/ On 2026년 10월 15일, Anthropic announced that it is providing employee-like access to the METR team for safety evaluation.
2/ The notice is also listed with its date in the latest items on the Anthropic newsroom, and around the same period the responsible scaling policy roadmap also continues the trend of expanding external evaluation and verification.
3/ This announcement is significant in that it allows an external evaluator to verify the system based on deeper access.
4/ It can be seen as an approach to checking safety in a way that is closer to the actual operational environment than ordinary public testing. [1][2]
Implications
5/ Providing employee-like access to an independent evaluator can improve the effectiveness of safety validation and better reveal vulnerabilities and operational risks under real usage conditions.
6/ In materials and science fields, auditability and verification systems for models applied to high-reliability research assistance, literature mining, experimental automation, and agentic workflows may become even more important.
7/ In addition, as external evaluation strengthens, institutions and research communities may need to review not only performance but also access privileges, data boundaries, and security controls when adopting models. [1][2]