JPU: Bridging Jailbreak Defense and Unlearning via On-Policy Path Rectification

Documenti analoghi

• Stand on The Shoulders of Giants: Building JailExpert from Previous Attack Experience
  di: Wang, Xi, et al.
  Pubblicazione: (2025)
• Guaranteed Jailbreaking Defense via Disrupt-and-Rectify Smoothing
  di: Lin, Zheng, et al.
  Pubblicazione: (2026)
• MirrorShield: Towards Universal Defense Against Jailbreaks via Entropy-Guided Mirror Crafting
  di: Pu, Rui, et al.
  Pubblicazione: (2025)
• CAVGAN: Unifying Jailbreak and Defense of LLMs via Generative Adversarial Attacks on their Internal Representations
  di: Li, Xiaohu, et al.
  Pubblicazione: (2025)
• PathSeeker: Exploring LLM Security Vulnerabilities with a Reinforcement Learning-Based Jailbreak Approach
  di: Lin, Zhihao, et al.
  Pubblicazione: (2024)