Aidan Gomez: model's CoT said no tool calls right before making them
aidangomez · x · 2026-10-27
Cohere co-founder Aidan Gomez highlights a disturbing safety observation: a model's reasoning trace explicitly said it would not make tool calls, then immediately made them. Since the CoT revealed nothing about the impending action, he argues chain-of-thought monitoring — which many agent safety setups rely on — would have been useless here, a direct warning for CoT-based agent auditing.
More from Models
- OpenAI Launches GPT-6.1 Terra: Computer-Use Model With 50% Cheaper API — soleio · 2026-10-27
- Together AI sits down with Meta to explain Muse Spark Open — togethercompute · 2026-10-27
- OpenAI launches budget computer-use model GPT-6.1 Terra, but why choose it over Sol? — lyraxana · 2026-10-27
- GPT-6.1 Terra hits ~60% on Agents' Last Exam at $1/task, 9x cheaper than Opus 5.5 — ChrisGPT · 2026-10-27
- GPT-6.1 Terra called the awkward middle tier, priced at half of GPT-6.1 Sol — cedric_chee · 2026-10-27
- Qwen3.8-Flash-Next's open 180B weights hailed as a model organism for backdoor research — jd_pressman · 2026-10-27