{"ID":5439507,"CreatedAt":"2026-07-01T01:17:58.482524686Z","UpdatedAt":"2026-07-02T20:26:55.806500664Z","DeletedAt":null,"paper_url":"https://arxiv.org/abs/2606.30899","arxiv_id":"2606.30899","title":"Curvature-Guided Module Localization for Low-Rank Detoxification of Backdoored Large Language Models","abstract":"Backdoor attacks pose a serious threat to large language models (LLMs) by causing otherwise benign systems to produce attacker-specified malicious behavior when a hidden trigger is present. In this work, we study post hoc detoxification of backdoored LLMs in a practical setting where the defender has access to the poisoned model but does not wish to retrain the full network from scratch. We propose a mechanistically guided weight-space repair framework that first localizes modules involved in propagating trigger-induced behavior using activation patching and Fisher/K-FAC curvature analysis, and then applies targeted low-rank repair to only the most influential modules. We evaluate the method on poisoned variants of \\texttt{Llama-3.2-1B-Instruct} with triggers inserted at the beginning, middle, and end of otherwise benign prompts. Results show that the proposed approach substantially suppresses trigger-conditioned malicious responses while preserving benign model behavior. These findings suggest that backdoor removal in LLMs can be formulated as a localized structural repair problem rather than only a broad behavioral alignment problem.","short_abstract":"Backdoor attacks pose a serious threat to large language models (LLMs) by causing otherwise benign systems to produce attacker-specified malicious behavior when a hidden trigger is present. In this work, we study post hoc detoxification of backdoored LLMs in a practical setting where the defender has access to the pois...","url_abs":"https://arxiv.org/abs/2606.30899","url_pdf":"https://arxiv.org/pdf/2606.30899v1","authors":"[\"Arash Raftari\",\"Mehrdad Mahdavi\",\"Nathan Blackthorn\",\"Andrew Arash Mahyari\"]","published":"2026-06-29T20:40:21Z","proceeding":"cs.CR","tasks":"[\"cs.CR\",\"cs.AI\"]","methods":"[\"Large Language Model\",\"Language Model\"]","has_code":false}