Beyond Attack Success Rate: Temporal Logit Observability for LLM Safety Failures

Documenti analoghi

• Persona Attack: Incremental Memory Injection Jailbreak Attack against Large Language Models
  di: Park, Junyoung, et al.
  Pubblicazione: (2026)
• ASGuard: Activation-Scaling Guard to Mitigate Targeted Jailbreaking Attack
  di: Park, Yein, et al.
  Pubblicazione: (2025)
• CyBiasBench: Benchmarking Bias in LLM Agents for Cyber-Attack Scenarios
  di: Lim, Taein, et al.
  Pubblicazione: (2026)
• Hierarchical Vision Language Action Model Using Success and Failure Demonstrations
  di: Park, Jeongeun, et al.
  Pubblicazione: (2025)
• Expanding Search Space with Diverse Prompting Agents: An Efficient Sampling Approach for LLM Mathematical Reasoning
  di: Lee, Gisang, et al.
  Pubblicazione: (2024)