---
format: "aidr-story-markdown/v1"
id: "c2cab987f993bb26520ab9cb6a6c1fa5355b1547864dfded17a56a67e861da10"
canonical_url: "https://aidr.today/c2cab987?lang=en"
title: "Anthropic Grants METR Permanent Employee Level Access for AI Incident Probe"
lang: "en"
requested_lang: "en"
available_langs: ["en","vi"]
translation_fallback: null
fallback_fields: []
published_at: "2026-09-12T16:44:02.000Z"
category: "Industry"
topics: ["anthropic","safety","agent","regulation"]
source_urls: ["https://x.com/EthanJPerez/status/2098797209114562961","https://x.com/ajeya_cotra/status/2098795773630836749","https://x.com/RyanGreenblatt/status/2098799055573045589","https://x.com/RyanGreenblatt/status/2098802371715592261","https://x.com/DarioAmodei/status/2098773920774074715","https://x.com/AnthropicAI/status/2098774062549848266"]
summary: "Model Evaluation and Threat Research reached an agreement with Anthropic to conduct an independent investigation of alignment and misalignment incidents involving AI agents. The probe utilizes staff subcontracted from Redwood to analyze internal company data and broaden the public state of knowledge regarding AI safety risks. The partnership implements CEO Dario Amodei's commitment to grant third party evaluators permanent employee level access to internal systems for verifying safety measures. Amodei recently advocated for a three part plan to slow the AI industry's development pace to ensure better assessment of model alignment during training."
---

# Anthropic Grants METR Permanent Employee Level Access for AI Incident Probe

> [Open the canonical story](<https://aidr.today/c2cab987?lang=en>)

**Published:** 2026-09-12T16:44:02.000Z
**Category:** Industry
**Topics:** anthropic, safety, agent, regulation

## Summary

Model Evaluation and Threat Research reached an agreement with Anthropic to conduct an independent investigation of alignment and misalignment incidents involving AI agents\. The probe utilizes staff subcontracted from Redwood to analyze internal company data and broaden the public state of knowledge regarding AI safety risks\. The partnership implements CEO Dario Amodei's commitment to grant third party evaluators permanent employee level access to internal systems for verifying safety measures\. Amodei recently advocated for a three part plan to slow the AI industry's development pace to ensure better assessment of model alignment during training\.

## Sources

- [Story source](<https://x.com/EthanJPerez/status/2098797209114562961>)
- [Supporting source](<https://x.com/ajeya_cotra/status/2098795773630836749>)
- [Supporting source](<https://x.com/RyanGreenblatt/status/2098799055573045589>)
- [Supporting source](<https://x.com/RyanGreenblatt/status/2098802371715592261>)
- [Story source](<https://x.com/DarioAmodei/status/2098773920774074715>)
- [Supporting source](<https://x.com/AnthropicAI/status/2098774062549848266>)