@article{agentlensinterpretablesafetysteeringviam, title = {AgentLens: Interpretable Safety Steering via Mechanistic Subspaces for Multi-Turn Coding Agent}, author = {Weidi Luo and Qiming Zhang and Yihao Quan and Mingyu Jin and Jie Cai and Chaowei Xiao and Jingcheng Niu and Zhen Xiang}, year = {2026}, eprint = {2606.22673}, archivePrefix = {arXiv}, url = {https://arxiv.org/abs/2606.22673}, }