arXivYitong Guo, Xiaoyi Chen, Siyuan Zhang, Xiaofeng Wang, Haixu TangTue, Sep 1, 2026, 8:59 AM PDT
score 16.5
Why Fine-Tuning AI Models Silently Disables Their Safety Features
Original: When Safety Routing Breaks: Understanding Alignment Fragility under Benign Fine-Tuning
Source: arxiv.org ↗
Writing ELI5 summary…