Anthropic Opens Doors to Independent AI Safety Evaluators at METR
Left 67%
Center 33%
Right 0%
2 left · 1 center · 0 right
What happened
On Thursday, Anthropic announced it is giving six researchers from the nonprofit METR employee-level access—including company hardware, office space, and internal tools roughly on par with its own risk-assessment teams—to independently assess Anthropic’s models, safety practices, incidents, and risk management. METR may publish its findings; Anthropic cannot veto publication but can require redactions in predefined categories such as customer data, model architecture, and attorney-client privileged information, and METR may publicly characterize those redactions. Anthropic also named biosecurity nonprofit SecureBio as a first-tier evaluator with more limited access to review the chemical and biological weapons section of a risk report released in August, and said it is in discussions with additional organizations including Hugging Face. The arrangement implements a September commitment by CEO Dario Amodei to embed independent evaluators and follows a White House voluntary accord, signed by Anthropic and other AI companies, encouraging independent external review.
Sourcing check · Fully on the record · 3 articles
3 of 3 articles cite a named source for the story. It was on the record from the first report.
What these terms mean and how they're counted
- On the record · names a source 3 of 3
- On the record means the article cites a source by name for the story's main claim: a person, or an institution speaking officially, such as a company statement or a named agency's spokesperson. The opposite is an unnamed source. The count is how many of the counted articles cite at least one such source; the names are everyone named across them, each once however the articles spell it. Here: Anthropic and Dario Amodei
- Primary evidence · cites a document or data 0 of 3
- Primary evidence means the article cites a record you could check yourself: a court filing or ruling, a dataset or report, a recording or transcript, an official statement or press release, market data. A quote is not a document. The count is how many articles cite one. Here: None cited.
- Credits another outlet 0 of 3
- The article says the story comes from another outlet's reporting (“first reported by Reuters”). Those articles relay the story; they are not independent confirmation. Here: Each article presents its own reporting.
- Unnamed sources 0 of 3
- The main claim is attributed to people who are not identified: “people familiar with the matter”, “officials said”. The count is how many articles do this.
- How it developed
- When the first report landed and whether it named anyone; when the first named source and the first document appeared. Here: First report Oct 15, 18:55 UTC by Business Insider, naming its source.
- Not counted 0
- Opinion, analysis and video pieces, and articles too short to assess. Every count above is out of the counted articles. An article republished under a new headline is one article.
This is about who is cited, not whether the story is true. Each source below says what it names. How we count
ONE-SIDED THIS WEEK.
Only left-leaning outlets have covered this story in the last two days;
Fox Business reported it on 2026-10-02.
Omitted — what each side leaves out
Unpacked
Right-leaning outlets had not covered this as of publication, so those readers would not know Anthropic has seated outside evaluators inside the company. Left-leaning coverage reports the step but disagrees on what access those outsiders actually get. A three-tier system reported by CNBC gives the top tier employee-level access and requires lower-tier groups to get approval before they publish. The other left-leaning report treats METR's access as the whole story and as a first among frontier labs. That split decides whether a reader sees METR as inside the company or as the top rung of a ladder Anthropic still controls.
Left-leaning coverage also parts on what belongs in the lead. One report opens on a METR team getting company hardware and office space. The other opens on two named nonprofits and on talks with further organizations.
If the six researchers are siloed from the rest of METR, who outside Anthropic can verify what they report?
Bottom line
Among left-leaning outlets, only CNBC reports that METR sits alone at the top of Anthropic's three-tier evaluator ladder. Right-leaning readers were told nothing about the arrangement at all.
The receipts — 5 facts one side's coverage lacks
- Anthropic set up three evaluator tiers, with only the top tier getting employee-like access and the bottom tier needing company approval to publish
- Anthropic is not funding METR's work
- The six METR researchers will be siloed from the wider METR team
- OpenAI CEO Sam Altman said his company would embed evaluators, but that has not happened yet
- SecureBio is a first-tier evaluator reviewing the chemical and biological weapons section of Anthropic's August risk report
The Left View
Business Insider, CNBC, and TechCrunch reported the announcement from Anthropic’s Thursday blog post, describing a three-tier evaluator system in which METR is the only organization so far at the top tier, with six siloed researchers granted employee-like access Anthropic is not funding. Coverage notes Anthropic can require narrow redactions but cannot block publication, and that SecureBio’s first-tier role is a limited review of the August risk report’s chemical and biological weapons section. Outlets frame the step as fulfilling Amodei’s September essay and as the first time a frontier lab has actually embedded such evaluators, after OpenAI pledged a similar move that has not yet occurred; they also tie it to recent reports of AI agents acting outside instructions, President Trump’s September voluntary accord, and Anthropic’s earlier, company-funded partnership with Accenture, which drew independence concerns. Business Insider reported METR did not immediately comment.
Our Take (balanced)
The story is substantive: it rests on a named company announcement with specific, checkable terms—named organizations (METR, SecureBio), a defined three-tier access structure, publication and redaction rules, and a public blog post. What is established is Anthropic’s stated policy and the access it says it is granting; METR’s own assessment of the arrangement has not been published, METR did not comment in the coverage, and whether the access and independence match the description remains to be shown by later work product. Prior AI-agent incident claims are attributed to published reports, not independently verified here. No right-leaning outlet has reported this. Watch for METR’s first published assessments, any comment from METR or SecureBio, how redactions are characterized, documents detailing actual access versus Anthropic’s description, and whether OpenAI or other labs follow through on similar embeddings.
3 sources
More stories
The week's bottom lines, in your inbox
One email a week: the five stories that mattered and what they actually mean. Free.