作者位次 #3/8ACL 2026Poster· Natural Language Processing
TROJ ail: Trajectory-Level Optimization for Multi-Turn Large Language Model Jailbreaks with Process Rewards
Xiqiao Xiong · Ouxiang Li · Zhuo Liu · Moxin Li · Wentao Shi · Fengbin ZHU · Qifan Wang · Fuli Feng
摘要
Large language models have seen widespread adoption, yet they remain vulnerable to multi-turn jailbreak attacks, threatening their safe deployment. This has led to the task of training automated multi-turn attackers to probe model safety vulnerabilities. However, existing approaches typically rel...