COSMIC: Generalized Refusal Direction Identification in LLM Activations

Documenti analoghi

• RepIt: Steering Language Models with Concept-Specific Refusal Vectors
  di: Siu, Vincent, et al.
  Pubblicazione: (2025)
• Peer-Preservation in Frontier Models
  di: Potter, Yujin, et al.
  Pubblicazione: (2026)
• SteeringSafety: A Systematic Safety Evaluation Framework of Representation Steering in LLMs
  di: Siu, Vincent, et al.
  Pubblicazione: (2025)
• Agent Instructs Large Language Models to be General Zero-Shot Reasoners
  di: Crispino, Nicholas, et al.
  Pubblicazione: (2023)
• A Framework for Formalizing LLM Agent Security
  di: Siu, Vincent, et al.
  Pubblicazione: (2026)