Shared Latent Structures Enable Unified Backdoor Detection and Mitigation in LLMs 文章

ArXiv CS.AI2026-06-09NEWSen作者: Omar Mahmoud, Aly M. Kassem, Thommen George Karimpanal, Buddhika Laknath Semage, Negar Rostamzadeh, Golnoosh Farnadi, Santu Rana
查看原文 →
关系图谱

Shared Latent Structures Enable Unified Backdoor Detection and Mitigation in LLMs · 相关技术

相关技术

Sparse Autoencoder (SAE)fine-tuningbidirectional activation steeringweight-editing attackscountry-conditioned harmful advicesentiment misclassificationbias inductionpassword-lockingrefusal manipulationjailbreakingresidual-stream activationslarge language models