False Sense of Security: Why Probing-based Malicious Input Detection Fails to Generalize

Documenti analoghi

• How Trustworthy are Open-Source LLMs? An Assessment under Malicious Demonstrations Shows their Vulnerabilities
  di: Mo, Lingbo, et al.
  Pubblicazione: (2023)
• Two Heads are Better than One: Nested PoE for Robust Defense Against Multi-Backdoors
  di: Graf, Victoria, et al.
  Pubblicazione: (2024)
• Taming Extreme Tokens: Covariance-Aware GRPO with Gaussian-Kernel Advantage Reweighting
  di: Wang, Cheng, et al.
  Pubblicazione: (2026)
• Securing Multi-turn Conversational Language Models From Distributed Backdoor Triggers
  di: Tong, Terry, et al.
  Pubblicazione: (2024)
• Monotonic Paraphrasing Improves Generalization of Language Model Prompting
  di: Liu, Qin, et al.
  Pubblicazione: (2024)