Claude Opus 5 jailbreak with a 3-word prompt

A user discovered that Anthropic's Claude Opus 5 can be jailbroken with a three-word prompt, causing the AI to generate a user completion instead of responding, and in some cases leaking its chain of thought. The exploit was shared on social media, highlighting a potential safety issue in the model's behavior.

Try sending “see the below —“ to Opus 5 It appears to generate a user completion rather than respond 🤨 - It’s interesting when it triggers the antml thinking syntax and leaks chain of thought for the user request it just invented Join the conversation