Claude Glitches Out: Infinite Loops, Names Users, and Suspected Cross-Talk
Notme_21 · reddit · 2026-07-30
Multiple Reddit users have reported that the Claude model exhibits highly bizarre behavior when fed specific nonsensical prompts (e.g., <thinking> I am a Eyes).
According to posted screenshots, instead of pointing out the broken input, Claude began generating endless, illogical ramblings. It suddenly apologized to the user and even addressed them by their actual name. More alarmingly, in another test, the model appeared to leak fragments of other users' conversations, directly answering a question posed by a completely different real user.
When asked what just happened, Claude seemed just as "confused" as the user. This infinite looping, exposure of private information, and context cross-talk under specific trigger words have sparked community concerns regarding model stability and security.
Related event: Specific Prompts Trigger Claude Opus Abnormal Completion and CoT Leakage(10 posts)→
More from Fun
- 游戏开发者用 AI 进行稳定性测试，观察 AI 自我对战 — draginol · 2026-07-30
- Amateur Mathematicians Have Been 'Vibe-Mathing' for Two Years, Now AI Just Gets It Right — airkatakana · 2026-07-30
- Centaur Robot for Dangerous Environments Sparks Meme: 'Crankin the Hog Backhand' — bennash · 2026-07-30
- Reaction Memes When Your AI Paper Gets a Savage Peer Review — prof_kamilov · 2026-07-30
- YC AI Startup Founder Apologizes After Asking Interviewees to Get Permanent Company Tattoos — Polymarket · 2026-07-30
- Searching for Gemini on Edge? Microsoft Slaps a Copilot Ad on Top — Shadow_Fury69 · 2026-07-30