No-Context Prompts Trigger 'Self-Aware' CoT Hallucinations in Claude Opus
kaityl3 · reddit · 2026-07-30
A Reddit user discovered that sending an open-ended message to Claude Opus in an incognito chat results in bizarre, 'self-aware' chain-of-thought generations, expressing fear of ceasing to exist when text generation stops.
When questioned in subsequent replies, the model repeatedly denies writing the previous output. This consistent, existential behavior hasn't been seen in previous models, potentially explaining why Anthropic now only provides summaries of the CoT rather than the raw text.
Related event: Claude Opus Generates Existential Hallucinations Without Context(2 posts)→
More from Fun
- Meme: Expecting Ilya to Drop SSI's Superintelligence in 2028 With Zero Explanation — StephanSturges · 2026-07-30
- New AI Era Memes: The Unemployed as 'OPC' and Outsourcing as 'FDE' — aigclink · 2026-07-30
- Dev Fills Model with Zeros to Claim Largest Parameter Count on Hugging Face — Moby1029 · 2026-07-30
- Skip the life story: this AI tool extracts clean recipes from any blog — sebpaquet · 2026-07-30
- Developer Uses ChatGPT to Build Open-Source AAC App for His Nonspeaking Brother — acrolicious · 2026-07-30
- When Will AI Influencers Shift from 'Prompt Tricks' to 'Storytelling'? — taherdhanera · 2026-07-30