Key points
- Anthropic gave six METR researchers company laptops, office space and internal tools roughly on par with its own risk assessment teams, per its Thursday blog post.
- METR can publish its findings without Anthropic's approval; Anthropic may require redactions only for narrow categories such as customer data, privileged material and model architecture.
- The deal is the top of a three-tier system; SecureBio, Accenture and Faculty get less access and tighter publishing limits, and talks with Hugging Face have begun.
- The move follows a White House accord on independent review signed by Anthropic, OpenAI, Google, Meta, Nvidia and xAI, per TechCrunch.
Why it matters for marketers: Marketers building on frontier models may soon see independent, publicly published assessments of vendor risk, giving procurement and brand-safety teams something firmer than vendor claims to cite.
Anthropic said Thursday that six researchers from METR, a nonprofit that tests frontier AI models, now have employee-level access to assess its models, safety practices and risk management. They have company laptops, office space and internal tools roughly comparable to what Anthropic's own risk teams use.
METR will publish its own assessments, and Anthropic has given up the right to veto them. It can require redactions only in narrow, predefined categories, including customer data, attorney-client privileged material and model architecture details. METR may say publicly whether it thinks those redactions were justified. Anthropic compared the setup to regulators placing resident examiners inside large banks.
METR sits at the top of a three-tier system. Reviewers such as SecureBio, Accenture and Faculty get less access and need approval to publish, according to The Decoder and TechCrunch. Anthropic has also started early talks with Hugging Face. TechCrunch reports the step follows a White House accord on independent review signed by several major AI developers, and comes amid recurring disclosures of AI agents acting outside their instructions.
For marketing and content teams, this matters most if you run agents or generation workflows on Claude. Independent reports on incidents and risk practices could become a reference point in vendor reviews, client contracts and internal policy, replacing reliance on company statements alone.
Watch for METR's first published assessment, and for how often redactions appear and how METR characterizes them. Also note whether other labs offer similar access, since that would make comparisons between vendors easier.
Sources
This brief is our own summary. Read the full reporting at the source.
