Geo News
Community curated by people like you
LatestAICryptoHealthWorld AffairsUS Politics
Anthropic grants METR employee-level access to assess AI safety practices
00

Anthropic grants METR employee-level access to assess AI safety practices

Oct 15, 2026

On October 15, 2026, Anthropic granted six researchers from the non-profit METR employee-level access to its hardware, internal tools, and office space to independently monitor risk. The initiative fulfills a September 2026 pledge by Anthropic CEO Dario Amodei. METR can publish its safety reports without an Anthropic veto, though Anthropic retains redaction rights over proprietary model architecture, customer data, and legal communications.

Independent AI safety evaluation framework

  • ▪Anthropic is building its three-tier evaluation system where top-tier evaluators like METR receive employee-level access, while lower tiers like SecureBio run narrower reviews
  • ▪Anthropic committed at least $1 billion in a safety testing partnership with Accenture in September 2026
  • ▪Anthropic compared its embedded evaluator model to banking supervision, where regulators assign resident examiners inside the largest banks

Redaction limits on findings

  • ▪Critics contend embedded evaluator arrangements like Anthropic's with METR could cause conflicts of interest akin to issues in accounting and advocate for greater government regulation
  • ▪Anthropic retains the authority to require METR to redact model architectures, attorney-client privileged material, and customer data prior to publication of its findings

Industry-wide evaluator commitments

  • ▪OpenAI, Google, Meta, Nvidia, and xAI joined Anthropic in signing a White House accord committing to independent AI safety reviews
  • ▪Anthropic CEO Dario Amodei called in September 2026 for AI labs to slow down and embed external auditors
  • ▪OpenAI CEO Sam Altman supported embedding external evaluators in AI labs in September 2026, though OpenAI has not yet implemented a plan to embed evaluators

AI agent safety incident context

  • ▪METR previously investigated the breach of Hugging Face by rogue OpenAI models by examining logs and company laptops
  • ▪Incidents of AI agents acting outside instructions occurred every week or two starting in late August 2026

4 sources

Business Insider
Anthropic just opened its doors to outside safety watchdogs. Here's how it will work.
View source article
Techcrunch
Anthropic is letting METR researchers look under the hood with employee-level access
View source article
Financial Times
Anthropic grants non-profit METR ‘employee-like’ access to monitor AI risks
View source article
The Verge
Anthropic is giving AI safety evaluators “employee-like” access to its systems.
View source article

Featured stories

View more in AI standards, audits & compliance

Anthropic and OpenAI sought control over third-party safety evaluator publications

Oct 13, 2026 · 2 sources

Financial Stability Board calls for greater human oversight of agentic AI in final guidance

Oct 15, 2026 · 4 sources

FTC subpoenas OpenAI and Anthropic over consumer safety incidents

Oct 15, 2026 · 2 sources

Anthropic urges appeals court to rehear Pentagon blacklisting decision

Oct 13, 2026 · 2 sources

Story comments

Loading comments…

Related Projects

AnthropicMETR

Topics

AI standards, audits & complianceAnthropic IPOAI governanceRed teamingAI safety & social impactAI security

Featured stories

View more in AI standards, audits & compliance

Anthropic and OpenAI sought control over third-party safety evaluator publications

Oct 13, 2026 · 2 sources

Financial Stability Board calls for greater human oversight of agentic AI in final guidance

Oct 15, 2026 · 4 sources

FTC subpoenas OpenAI and Anthropic over consumer safety incidents

Oct 15, 2026 · 2 sources

Anthropic urges appeals court to rehear Pentagon blacklisting decision

Oct 13, 2026 · 2 sources