mlsanityData to 14 Oct 2026

ASRU: Activation Steering Meets Reinforcement Unlearning for Multimodal Large Language Models

arXiv 2605.15687 ↗First posted 15 May 2026cs.CL cs.AIcited by 1

On the map: Multimodal LLMs / vision-language models → MLLM jailbreaks and adversarial robustness

also in 2 more
Abstract (arXiv)

Multimodal large language models (MLLMs) may memorize sensitive cross-modal information during pretraining, making machine unlearning (MU) crucial. Existing methods typically evaluate unlearning effectiveness based on output deviations, while overlooking the generation quality after unlearning. This can easily lead to hallucinated or rigid responses, thereby affecting the usability and safety of the unlearned model. To address this issue, we propose ASRU, a controllable multimodal unlearning framework that incorporates generation quality as a core evaluation objective. ASRU first induces initial refusal behavior through activation redirection, and then optimizes fine-grained refusal boundaries using a customized reward function, thereby achieving a better trade-off between target knowledge unlearning and model utility. Experiments on Qwen3-VL show that ASRU significantly improves unlearning effectiveness (+24.6%) on average and generation quality (5.8X) on average while effectively preserving model utility, using only a small amount of retained supervision data.

Key facts. Cited by 1 paper in the map (data through 14 Oct 2026). Placed in MLLM jailbreaks and adversarial robustness, one of 16 directions in Multimodal LLMs / vision-language models. Builds on 19 of its 45 references that are in the map. First posted on arXiv 15 May 2026.

Builds on

19 of 45 references.

Closest ancestors · 3

Same direction, newest first

  1. 2025MLLMEraser: Achieving Test-Time Unlearning in Multimodal Large Language Models through Activation Steering5 Oct
  2. 2025Activation Steering Meets Preference Optimization: Defense Against Jailbreaks in Vision Language Models30 Aug
  3. 2025SafeEraser: Enhancing Safety in Multimodal Large Language Models through Multimodal Machine Unlearning18 Feb

Other references · 16

  1. 2024DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Modelscited by 9,379
  2. 2024Qwen2-VL: Enhancing Vision-Language Model's Perception of the World at Any Resolutioncited by 4,234
  3. 2023Steering Language Models With Activation Engineeringcited by 922
  4. 2023The Linear Representation Hypothesis and the Geometry of Large Language Modelscited by 716
  5. 2023Safe RLHF: Safe Reinforcement Learning from Human Feedbackcited by 656
  6. 2024TOFU: A Task of Fictitious Unlearning for LLMscited by 533
  7. 2024Negative Preference Optimization: From Catastrophic Collapse to Effective Unlearningcited by 504
  8. 2024Improving Instruction-Following in Language Models through Activation Steeringcited by 160
  9. 2024Towards Safer Large Language Models through Machine Unlearningcited by 139
  10. 2023Improving Activation Steering in Language Models with Mean-Centringcited by 68
  11. 2025Rethinking LLM Unlearning Objectives: A Gradient Perspective and Go Beyondcited by 61
  12. 2024Representation Surgery: Theory and Practice of Affine Steeringcited by 50
4 more
  1. 2025AlphaSteer: Learning Refusal Steering with Principled Null-Space Constraintcited by 49
  2. 2023Visual Instruction Tuning towards General-Purpose Multimodal Model: A Surveycited by 29
  3. 2025RULE: Reinforcement UnLEarning Achieves Forget-Retain Pareto Optimalitycited by 16
  4. 2025Praxis-VLM: Vision-Grounded Decision Making via Text-Driven Reinforcement Learningcited by 9

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2026: 1.

1 paper.

  1. 2026CUSE: Fine-Grained Multimodal Unlearning with Counterfactual Guidance and On-Policy Self-Distillationcited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2026Visual-Noise Guided In-Context Distillation for Multimodal Large Language Model Unlearning
  2. 2024Rethinking Machine Unlearning for Large Language Models
  3. 2026CAP: Controllable Alignment Prompting for Unlearning in LLMs
  4. 2025FALCON: Fine-grained Activation Manipulation by Contrastive Orthogonal Unalignment for Large Language Model
  5. 2024Machine Unlearning of Pre-trained Large Language Models
  6. 2026Are We Really Measuring Forgetting? Pitfalls in Evaluating Machine Unlearning for Multimodal LLMs
  7. 2026A Comprehensive Evaluation of LLM Unlearning Robustness under Multi-Turn Interaction
  8. 2024Towards Effective Evaluations and Comparisons for LLM Unlearning Methods