When Safety Becomes An Outlier: Understanding the Retention of LLM Safety Behaviors
cs.CL
Submitted: 2026-11-07
Updated: 2026-11-07
Code: https://github.com/tatsu-lab/stanford_alpaca
Terminology
Sources
- Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback
- Constitutional AI: Harmlessness from AI Feedback
- Safety-Aware Fine-Tuning of Large Language Models
- Training Verifiers to Solve Math Word Problems
- The Llama 3 Herd of Models
- Deliberative Alignment: Reasoning Enables Safer Language Models
- Harmful Fine-tuning Attacks and Defenses for Large Language Models: A Survey
- Mistral 7B
- LoRA Fine-tuning Efficiently Undoes Safety Training in Llama 2-Chat 70B
- Trustworthy LLMs: a Survey and Guideline for Evaluating Large Language Models' Alignment
- Proximal Policy Optimization Algorithms
- An Embarrassingly Simple Defense Against LLM Abliteration Attacks
- DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models
- Do-Not-Answer: A Dataset for Evaluating Safeguards in LLMs
- Qwen2.5 Technical Report
- AsFT: Anchoring Safety During LLM Fine-Tuning Within Narrow Safety Basin
- Shadow Alignment: The Ease of Subverting Safely-Aligned Language Models
- Universal and Transferable Adversarial Attacks on Aligned Language Models
Related papers
- ALARM: Audio-Language Alignment for Reasoning Models
- ThinkSum: Probabilistic reasoning over sets using large language models
- Coherence boosting: When your pretrained language model is not paying enough attention
- From Facts to Insights: A Persona-Driven Dual Memory Framework and Dataset for Role-Playing Agents
- Improving General Role-Playing Agents via Psychology-Grounded Reasoning and Role-Aware Policy Optimization
- The Art of Asking: Multilingual Prompt Optimization for Synthetic Data