Beyond Refusal Patterns: Safe-Role Internalization for Robust and Generalizable LLM Safety Alignment

arXiv:2610.07023 · cs.AI, cs.CL, cs.IR · Submitted 2026-10-04 · Read on arXiv

cs.AI, cs.CL, cs.IR

Submitted: 2026-10-04

Updated: 2026-10-04

Comments: 27 pages,7 figures, under review

Code: https://github.com/Code-PJH/SSRFT

License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/

Terminology

• Alignment — 1× in this paper · explained in 1 episode

Sources

• Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback
• Persona Vectors: Monitoring and Controlling Character Traits in Language Models
• Think you have Solved Question Answering? Try ARC, the AI2 Reasoning Challenge
• Machine Mindset: An MBTI Exploration of Large Language Models
• The Llama 3 Herd of Models
• An Empirical Study on Reinforcement Learning for Reasoning-Search Interleaved LLM Agents
• ChatHaruhi: Reviving Anime Character in Reality via Large Language Model
• Prefill-level Jailbreak: A Black-Box Risk Analysis of Large Language Models
• A Comprehensive Survey on Reinforcement Learning-based Agentic Search: Foundations, Roles, Optimizations, Evaluations, and Applications
• Do LLMs Possess a Personality? Making the MBTI Test an Amazing Evaluation for Large Language Models
• Comparing Traditional and LLM-based Search for Consumer Choice: A Randomized Experiment
• Gemma 2: Improving Open Language Models at a Practical Size
• Llama 2: Open Foundation and Fine-Tuned Chat Models
• A Survey of LLM-based Deep Search Agents: Paradigm, Optimization, Evaluation, and Challenges
• ChatGPT vs. Google: A Comparative Study of Search Performance and User Experience
• Qwen2.5 Technical Report
• Qwen3 Technical Report
• Any-Depth Alignment: Unlocking Innate Safety Alignment of LLMs to Any-Depth
• Universal and Transferable Adversarial Attacks on Aligned Language Models

Related papers

• MAVEN-T: Reinforced Heterogeneous Distillation for Real-Time Multi-Agent Trajectory Prediction
• Model Discovery Agent: LLM-assisted Bayesian experiment design for data-efficient discovery of mechanistic world models
• Agents' Last Exam
• The Clinician's Veto: Navigating Trust, Liability, and Uncertainty in Autonomous AI Prescribing
• MindHelper: Closed-Loop Embodied Mental-State Reasoning for Precision Intervention
• Incumbent Advantage: Brand Bias and Cognitive Manipulation Dynamics in LLM Recommendation Systems