---
format: "aidr-story-markdown/v1"
id: "8bf59036d175849047b5cbad4c28271c72bf157dfff1dc3899f808f2ee1f75e5"
canonical_url: "https://aidr.today/8bf59036?lang=en"
title: "Anthropic Offered Job to METR Auditor During AI Speedup Review"
lang: "en"
requested_lang: "en"
available_langs: ["en","vi"]
translation_fallback: null
fallback_fields: []
published_at: "2026-11-06T22:36:22.000Z"
category: "Research"
topics: ["anthropic","safety"]
source_urls: ["https://huggingnews.com/ai/anthropic-offered-job-to-metr-auditor-during-ai-speedup-review-cb62ba12","https://x.com/METR_Evals/status/2118412183860351084","https://x.com/METR_Evals/status/2118412205662314935","https://x.com/METR_Evals/status/2118412194702598339","https://x.com/Sam_McCandlish/status/2118422266669371519","https://x.com/ziv_ravid/status/2118439355442221072","https://x.com/choblin29/status/2118441015984185580","https://x.com/tobyordoxford/status/2118646665447936023"]
summary: "Anthropic offered a job to a METR auditor during an AI speedup review, a conflict of interest disclosed in METR's report that the staffer declined. The 100-page evaluation estimated benchmark progress on Anthropic's capability index accelerated 1.9x since October 2025, with internal AI use driving roughly 80% of the speedup. METR also analyzed three unreleased models trained between July and October 2026, including Model 4, which scored 178.6 on the capability index and led staff to self-report an average 12x productivity gain."
---

# Anthropic Offered Job to METR Auditor During AI Speedup Review

> [Open the canonical story](<https://aidr.today/8bf59036?lang=en>)

**Published:** 2026-11-06T22:36:22.000Z
**Category:** Research
**Topics:** anthropic, safety

## Summary

Anthropic offered a job to a METR auditor during an AI speedup review, a conflict of interest disclosed in METR's report that the staffer declined\. The 100\-page evaluation estimated benchmark progress on Anthropic's capability index accelerated 1\.9x since October 2025, with internal AI use driving roughly 80% of the speedup\. METR also analyzed three unreleased models trained between July and October 2026, including Model 4, which scored 178\.6 on the capability index and led staff to self\-report an average 12x productivity gain\.

## Sources

- [Story source](<https://huggingnews.com/ai/anthropic-offered-job-to-metr-auditor-during-ai-speedup-review-cb62ba12>)
- [Story source](<https://x.com/METR_Evals/status/2118412183860351084>)
- [Story source](<https://x.com/METR_Evals/status/2118412205662314935>)
- [Story source](<https://x.com/METR_Evals/status/2118412194702598339>)
- [Story source](<https://x.com/Sam_McCandlish/status/2118422266669371519>)
- [Supporting source](<https://x.com/ziv_ravid/status/2118439355442221072>)
- [Supporting source](<https://x.com/choblin29/status/2118441015984185580>)
- [Supporting source](<https://x.com/tobyordoxford/status/2118646665447936023>)