New RL-based method targets black-box adversarial attacks on language models
A new paper introduces RL-ATTACK, a reinforcement learning framework for generalized black-box adversarial attacks that perturb inputs to make large language models produce unintended responses.
Read the full story at arXiv cs.CL ↗
Timeline · 1 report
- 2026-11-09 05:00 · arXiv cs.CL
RL-ATTACK: A Generalized Black-Box Adversarial Attack on Language Models via Reinforcement Learning