Large reasoning models unlearn sensitive info without breaking thinking
REDaCT: Collapse-Free Unlearning in Large Reasoning Models
Machine LearningArtificial IntelligenceComputation and Language
Summary
Sometimes AI models have to forget certain facts, but this can make them stop reasoning clearly. The authors propose a new technique called REDaCT that spots where the AI might reveal forgotten info and fixes its thinking to avoid mistakes. This way, the AI forgets what it should but still makes sense and finishes its reasoning properly. Their tests show REDaCT does this better than earlier methods.
What this means in practice
- •For ai model engineers: Ensure AI systems can forget certain data without losing their ability to reason clearly during use.
- •For privacy compliance teams: Enable removal of sensitive or outdated information from AI models while keeping their responses coherent and complete.
Authors
Joshua Maccoby, Wendy Hui Wang
Abstract
Unlearning in large reasoning models (LRMs) requires removing forget-set information from both the reasoning trace and the final answer. Existing approaches achieve this objective at the cost of reasoning collapse, where the model fails to produce a coherent chain-of-thought (CoT) and does not properly terminate its reasoning process. To address this challenge, we propose REDaCT (REgenerate via Detection and Corrective Tuning), an inference-time unlearning framework for LRMs. REDaCT consists of two key components: (1) a binary leak detector that identifies sentences in the reasoning trace that leak forget-set information, and (2) an embedding optimization module that learns a corrective embedding offline, such that when injected at the first detected leak during inference, it guides subsequent reasoning towards sentences that do not leak forget-set information. Experimental results show that REDaCT completely avoids reasoning collapse, substantially improving reasoning preservation compared to state-of-the-art unlearning methods while maintaining perfect answer forgetting and near-perfect CoT forgetting. Our code is available at: https://github.com/jmaccoby/redact.