RL-ATTACK: A Generalized Black-Box Adversarial Attack on Language Models via Reinforcement Learning

Summary

The paper presents RL-ATTACK, a trainable black-box framework that uses reinforcement learning to generate adversarial transformations against language models without gradient access. On natural language inference, it reduces post-attack accuracy to 2% within 20 queries, and against generative models it beats an optimization-based baseline in 4 of 6 settings. The authors report that no reward design or training setting is consistently dominant.

Topics

• Adversarial machine learning

Related items

• InfoEvaluating the robustness of explainable AI methods to adversarial attacks in intrusion and malware detectionSimilar attack · OpenAlex (peer-reviewed AI security)
• InfoEvaluating Parameter-Efficient Fine-Tuning Attacks Against Non-Transferable Learning DefensesSimilar attack · Elsevier Security Journals
• InfoEscalate or Close: How Fast-Tier Models Handle Held-Out Web Attack Families in Two-Tier TriageSimilar attack · Arxiv (cs.CR + cs.CL + cs.LG)
• InfoVIPA: View-Invariant Projector-Based Adversarial AttackSimilar attack · Arxiv (cs.RO + cs.CV security)
• InfoAdversarial robustness of deep learning-based thyroid nodule segmentation in ultrasoundSimilar attack · OpenAlex (peer-reviewed AI security)