Neural Exec: Learning (and Learning from) Execution Triggers for Prompt Injection Attacks
Dario Pasquini, Martin Strohmeier, Carmela Troncoso · 2024
We introduce a new family of prompt injection attacks, termed Neural Exec. Unlike known attacks that rely on handcrafted strings (e.g., "Ignore previous instructions and..."), we show that it is possible to conceptualize the creation of execution triggers as a differentiable search problem and use learning-based methods to autonomously generate them.