Technology
Anthropic gives METR researchers employee-level access to assess its AI models
Six METR researchers will independently assess Anthropic's models and safety practices, and METR can publish its findings without Anthropic's approval.
The short version
- Anthropic is giving six researchers from the nonprofit METR employee-level access to independently assess its models, safety practices, and risk management.[TechCrunch]
- METR can publish its own assessments without Anthropic's veto, though Anthropic may require redactions in narrow categories such as customer data and model architectures.[TechCrunch]
- METR is currently the only evaluator at the top tier of a three-tier system Anthropic is building, while SecureBio sits in the first tier with narrower reviews that need Anthropic's approval to publish.[TechCrunch · CNBC]
Key facts
- Anthropic is giving six METR researchers employee-level access to independently assess its models, safety practices, and risk management.[TechCrunch]
- METR will publish its own assessments of Anthropic's risks, incidents, and practices, and Anthropic cannot veto that publication.[TechCrunch]
- Anthropic can require redactions for narrow categories, including customer data, attorney-client privileged material, and model architectures.[TechCrunch]
- METR is the only evaluator so far at the top tier of Anthropic's three-tier evaluator system.[TechCrunch]
- SecureBio is a first-tier evaluator with more limited access, and its lower-tier reviews require Anthropic's approval to publish.[TechCrunch · CNBC]
- Anthropic stated that it is not funding METR's work.[CNBC]
What remains uncertain
- How often Anthropic will use its redaction rights over METR's published findings remains unknown.[TechCrunch]
- The description of METR's independence and access comes from Anthropic's own announcement, and the reporting does not include independent confirmation of how the arrangement will work in practice.[TechCrunch · CNBC]
Sources
Outlet counts describe coverage, not independent confirmation. Reports may share a wire service or original source.