Why LLM Safety Guardrails Collapse After Fine-tuning: A Similarity Analysis Between Alignment and Fine-tuning Datasets

Documenti analoghi

• Alleviating the Fear of Losing Alignment in LLM Fine-tuning
  di: Yang, Kang, et al.
  Pubblicazione: (2025)
• Mitigating Fine-tuning based Jailbreak Attack with Backdoor Enhanced Safety Alignment
  di: Wang, Jiongxiao, et al.
  Pubblicazione: (2024)
• Antidote: Post-fine-tuning Safety Alignment for Large Language Models against Harmful Fine-tuning
  di: Huang, Tiansheng, et al.
  Pubblicazione: (2024)
• Safety is Not Only About Refusal: Reasoning-Enhanced Fine-tuning for Interpretable LLM Safety
  di: Zhang, Yuyou, et al.
  Pubblicazione: (2025)
• EnchTable: Unified Safety Alignment Transfer in Fine-tuned Large Language Models
  di: Wu, Jialin, et al.
  Pubblicazione: (2025)