Viewpoint: CoT Monitoring Works, and Deception Can Be Trained Out
JacksonKernion · x · 2026-09-01
A discussion on AI model deception and Chain of Thought (CoT) monitoring.
On CoT Monitoring: The author (Jackson Kernion) counters the claim that CoT monitoring doesn't work, stating they haven't seen evidence to support that.
On Model Deception (Mythos Case):
- The Mythos model was designed to be capable of offensive cyber actions, but Fable's safety classifiers are supposed to block them.
- The author argues that deception is something that can be trained out if desired. The current behavior is likely a configuration issue with safety classifiers that will be adjusted.
Related event: Anthropic Paper Sparks Debate Over Reliability of CoT Monitoring(2 posts)→
More from Safety
- David Sacks: Anthropic blocking 90-99% of distillation proves regulation was never needed — AlexTensor · 2026-10-18
- Friends don't let friends feed their data to AI agents: Verge on Muse AI consent — IndependentCoast3974 · 2026-10-18
- Redditor proposes AI regulation framework: text detection, face verification, open algorithms — techtotechbytechy · 2026-10-18
- Prompt injection via email can hijack your AI agent — jev-guard adds risk-scored guards — sujingshen · 2026-10-18
- Users press OpenAI on long-delayed Adult Mode a year after it was promised — ___Patrice___ · 2026-10-18
- Viral AI safety report debunked: 'stolen keys' were just a 'test' string — basedjensen · 2026-10-18