Type: url
Original: https://simonwillison.net/2024/Mar/5/prompt-injection-jailbreaking/
Status: generated
Summary
Simon Willison’s March 2024 argument that prompt injection and jailbreaking are distinct attack classes that are routinely conflated. Prompt injection depends on concatenating untrusted input with a trusted developer prompt and targets applications, while jailbreaking targets the safety filters built into the models themselves; he warns that a detection system trained on jailbreaks will not stop an application-specific injection, and reflects that a coined term needs ongoing maintenance.