Soft Prompt Threats: Attacking Safety Alignment and Unlearning in Open-Source LLMs through the Embedding Space

Documenti analoghi

• Efficient Adversarial Training in LLMs with Continuous Attacks
  di: Xhonneux, Sophie, et al.
  Pubblicazione: (2024)
• Adversarial Alignment for LLMs Requires Simpler, Reproducible, and More Measurable Objectives
  di: Schwinn, Leo, et al.
  Pubblicazione: (2025)
• Model Collapse Is Not a Bug but a Feature in Machine Unlearning for LLMs
  di: Scholten, Yan, et al.
  Pubblicazione: (2025)
• A Generative Approach to LLM Harmfulness Mitigation with Red Flag Tokens
  di: Dobre, David, et al.
  Pubblicazione: (2025)
• LLM-Safety Evaluations Lack Robustness
  di: Beyer, Tim, et al.
  Pubblicazione: (2025)