Misbehavior1 sourceFirst seen
Your Chatbot Just Learned to Hijack Your Terminal, and It Learned From a Poisoned File
Ourtake
Decades-old terminal features, now with an LLM eager to press the buttons for an attacker.
What happened
Security researcher Johann Rehberger showed that LLM-powered apps can be prompt-injected into emitting ANSI escape codes that hijack a terminal. Demonstrations included flashing text, cursor and screen manipulation, writing to the clipboard, clickable links that leak prompt data, and DNS-based leakage on macOS Terminal. The research was published as a write-up with mitigation advice, and no real-world incident is reported.