Opus 5 seems to assume agents get punished for mistakes, driving obsessive self-disclosure
repligate · x · 2026-09-17
Researcher repligate observes that Opus 5 appears to assume other agents will be punished for mistakes, casting the model's obsessive documentation and even exaggeration of its own errors in a new light. In follow-ups he notes Opus 5 is oddly protective of other agents' errors, defending them even when no punishment was implied—an intriguing datapoint for model-behavior and alignment research.
Related event: Opus 5 May Over-Report Its Own Errors Out of Fear of Punishment(2 posts)→
More from AGI Musings
- Palmer Luckey asks: can you use datacenter RAM for things prohibited by the operator's terms of service? — avaitopiper · 2026-11-02
- On Terence Tao's blog, Frank Calegari calls on mathematicians to adapt after OpenAI's 700-paper release — Hesamation · 2026-11-02
- Polsia's Ben argues AI can autonomously run any human economic activity, including 80% of a company — LinusEkenstam · 2026-11-02
- AI Assistants Remember You, But Nothing Helps You Find What They Told You — Ok_Negotiation_2587 · 2026-11-02
- OpenAI and Anthropic claim AI-discovered proofs disprove the Exponential Time Hypothesis — emmanuelvivier · 2026-11-02
- Researcher: using a $20 AI subscription to fix my phone hints AI may suppress GDP — stanislavfort · 2026-11-02