Thursday, October 15, 2026
Paperclip Index
Paperclip IndexDocumented harm17Negligible harm▼ 9 from a week agoThe Index

SignalSafety practiceSG-0039

Anthropic gave six METR evaluators insider access and cannot veto what they publish

The six work inside Anthropic with company hardware and internal tools on par with its own risk teams, and will publish findings on its risks and incidents. Anthropic can require redactions only in narrow categories such as customer data and model architecture, and METR may say publicly whether those were justified. SecureBio and Accenture work with more limited access under a new three-tier scheme, and early talks include Hugging Face.

Reported Oct. 15, 2026Signal: never counted in the index

Why it matters

An outside team with near-insider access and its own right to publish could bring to light incidents and risk decisions that a lab would otherwise describe only in its own words. This is how Anthropic is carrying out the evaluator pledge in the White House accord. The arrangement is voluntary and runs on agreements Anthropic set.

Sources

Read the reporting

1 source. Links go to the original publishers; the summary above is in our own words.

  1. primaryAn update on independent evaluatorsAnthropic · Oct. 15, 2026anthropic.com/news/an-update-on-independent-evaluators

Signals

More from the news desk

  1. Safety practice

    OpenAI opened GPT-Rosalind, its gated biology model, to US university labs at no cost

  2. Safety practice

    OpenAI used its own models to trace a hidden recording device in its offices to an employee, who was arrested

  3. Safety practice

    American Express launched a framework that checks AI agent purchases against what the customer authorized

All signals

Cite and share

Use this signal

Citation

Paperclip Index. “Anthropic gave six METR evaluators insider access and cannot veto what they publish.” Signal SG-0039. Reported Oct. 15, 2026; updated Oct. 15, 2026. https://paperclipindex.com/signal/SG-0039