1 entry with this tag.
A security vulnerability where malicious input causes an LLM to ignore its original instructions and execute unauthorized actions.