Skip to content
ML Academy

Introduced on day 53

Direct prompt injection

An attack in which the user's own input tells the model to ignore its instructions, change its goal or reveal its prompt.

In detail

Perez and Ribeiro (2022) describe goal hijacking and prompt leaking as the two main forms. Because the attacker is the user, direct injection mostly threatens the system prompt, the application's policy and the user's own session; its impact grows with the tools and data the session can reach.