Work

Neural Exec: Learning (and Learning from) Execution Triggers for Prompt Injection Attacks

Year: 2024

Type: conference-paper

Abstract: 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.

DOI: 10.1145/3689932.3694764

Authors: Dario Pasquini, Martin Strohmeier, Carmela Troncoso

Institutions: George Mason University, Bundesamt für Rüstung armasuisse, École Polytechnique Fédérale de Lausanne

Language: English

Topic: Adversarial Robustness in Machine Learning

Subfield: Artificial Intelligence

Field: Computer Science

Domain: Physical Sciences

Open Access status: green

OpenAlex ID: W4404612289

FWCI: 7.623

Cites: 15

Cited by: 26

Related to: 10