Towards Reasoning-Preserving Unlearning in Multimodal Large Language Models

TL;DR

This paper introduces RMLLMU-Bench, a benchmark for evaluating unlearning in multimodal large language models, and proposes R-MUSE, a method that effectively forgets sensitive information while preserving reasoning ability.

Contribution

The paper presents the first benchmark for reasoning-preserving unlearning in RMLLMs and a novel inference-time intervention method that balances forgetting with reasoning retention.

Findings

Existing methods either leak reasoning information or degrade reasoning ability.

R-MUSE outperforms baselines in balancing forgetting and reasoning retention.

Benchmark results highlight the need for specialized unlearning approaches.

Abstract

Machine unlearning aims to erase requested data from trained models without full retraining. For Reasoning Multimodal Large Language Models (RMLLMs), this is uniquely challenging: intermediate chain-of-thought steps can still leak sensitive information even when final answers are forgotten, and overly aggressive interventions easily damage general reasoning ability. Yet no benchmark jointly evaluates how well unlearning methods suppress reasoning-level leakage while preserving reasoning competence. We address this gap with RMLLMU-Bench, the first benchmark for RMLLM unlearning that extends standard forgetting metrics with dedicated measures of reasoning leakage and reasoning retention. A systematic evaluation on RMLLMU-Bench reveals that existing unlearning methods for MLLMs and Large (Language) Reasoning Models (LRMs) either leave substantial leakage in the reasoning process or severely degrade reasoning performance. To address these gaps, we propose R-MUSE (Reasoning-preserving MLLM Unlearning via Subspace guidance and Adaptive Steering), a training-free and inference-time intervention framework that steers internal representations to forget both answers and reasoning traces while explicitly preserving general reasoning. Experiments on RMLLMU-Bench demonstrate that R-MUSE achieves a substantially better balance between effective forgetting and reasoning retention.