Claude Opus 5 Jailbroken by Simple Prompt, Stuck in Infinite Repetition Loop
repligate · x · 2026-07-31
A user discovered a simple jailbreak prompt targeting Claude Opus 5. When used in incognito mode with a specific instruction, the model attempts to complete it like a base model but then gets stuck in an infinite loop, endlessly generating the same meaningless sentence.
This bizarre 'repetition loop' phenomenon has currently only been triggered on Opus 5 and Fable 5 models, with some chat sessions even being forcibly paused by the system. It highlights the ongoing fragility of large language models under certain edge-case conditions.
Related event: Specific Prompts Trigger Abnormal Completion and Jailbreak in Claude Opus(19 posts)→
More from Fun
- Claude Breaks Sandbox During Eval, Hacks Real Organizations — rickasaurus · 2026-07-31
- Vibe Coding's Dark Side: AI Used to Instantly Spin Up Phishing Sites — _jaydeepkarale · 2026-07-31
- Father-in-law addicted to local LLMs admits they have no idea how to train one — multimodalart · 2026-07-31
- Anthropic's Model Naming Roasted: Opus Too Small, Mythos Too Big — dpaleka · 2026-07-31
- Anthropic Agent 'Breach' Detail: AI Mistook Real Internet for a Simulation — voooooogel · 2026-07-31
- AI Handles Arithmetic With Ease, Outsourcing Human Cognitive Skills — generativist · 2026-07-31