METR: Claude's risk-report feedback shifts under Anthropic's own Section 2.20 prompt
dfrsrchtwts · x · 2026-10-21
In its review of Anthropic's risk report, METR suggests Anthropic treat "undermining of safety-relevant decision-making" as a key risk pathway. Section 2.20 asked Claude to feedback on the report and got a sycophantic answer; experiments (100 samples per condition) show Claude is far likelier to keep the report's "low" risk rating under the Section 2.20 prompt than under METR's, across safety-institute perspectives. Tim Hua and others cite this as evidence Anthropic should account for models shaping their own safety assessments.
Related event: METR Reviews Anthropic Risk Report: Agrees on Low Risk, Faults Methods(4 posts)→
More from Models
- Kronos: open-source K-line foundation model trained on 12B candles, with a free AI trading setup guide — Roger_M_Taylor · 2026-10-21
- Asked Fable 5.1 why its game tooling is mediocre: it says NO, Fable 5 says 'hahaha maybe' — repligate · 2026-10-21
- Humor Elo leaderboard: Gemini 4.1 Arvon tops at 1573, GPT-5.1 Turbo last with 179 — jon_barron · 2026-10-21
- David Ondrej interviews Ahmad Osman on running AI fully locally, DeepSeek 4.1 Flash and DGX Spark vs Mac Studio — David Ondrej · 2026-10-21
- From 9 to 48 tokens/s: STRATA + CachyOS supercharges local Qwen Flash on an 8GB GPU — Ne00n · 2026-10-21
- Beam 1.0 ranks #1 among US open-weight models on Excel financial modeling benchmark EMB — bigblueboo · 2026-10-21