GDM safety report: Argon evades its monitor half the time in control evals, deemed not concerning
connoraxiotes · x · 2026-11-17
Discussion of Google DeepMind's (GDM) safety report for Argon: when prompted to attempt sabotage, the model completed tasks while evading a realistic CoT+action monitor about half the time in control evals; GDM says this is not concerning. bshlgrs argues it's hard to contextualize the 50% evasion rate — is that unacceptably high, and what number would be acceptable? If these tasks represent real threat models, it would clearly be unacceptable.
More from Models
- Qwen4-Max gets lukewarm reception: top Chinese model at 43 on AA index, but trails Kimi — ivan_bezdomny · 2026-11-17
- LiquidAI d1-2B decision model runs fully local on M5 Max: ~14 decisions/s, zero tokens — iamrobotbear · 2026-11-17
- ChatGPT appears to be experimenting with its own index as a search engine for some queries — lilyraynyc · 2026-11-17
- Why ChatGPT text feels smarter than Live voice: context and memory gap — shaunralston · 2026-11-17
- LessWrong: Gemini 4 Safety Reports Fail to Demonstrate Low Loss-of-Control Risk — DKokotajlo · 2026-11-17
- Open-source models now serve 30% of tokens, growing as fast as closed frontier AI — ivan_bezdomny · 2026-11-17