mlsanityData to 7 Oct 2026

Toward Fine-Grained Forgetting:Attribute Unlearning for Multimodal Large Language Models

arXiv 2608.01008 ↗First posted 2 Aug 2026cs.AIcited by 0

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

also in 1 more
Abstract (arXiv)

Multimodal large language models (MLLMs) exhibit strong vision--language capabilities but may also memorize and disclose sensitive information. Machine unlearning seeks to remove designated knowledge without retraining from scratch while preserving general utility. Existing privacy-oriented benchmarks primarily adopt profile-level deletion, whereas practical requests are often finer grained: a model should forget a specified attribute while retaining non-sensitive information about the same identity. We therefore introduce attribute-level MLLM unlearning as a finer-grained task and construct a benchmark spanning long-text, numeric, and short-text targets, multiple forget ratios, and diverse question types. Our evaluation reveals that target and retained attributes share identity-specific and visual evidence, making selective forgetting susceptible to residual leakage or collateral degradation; accordingly, existing methods exhibit unstable forgetting--retention trade-offs in this setting. To address this challenge, we propose Causal Localization and Retain-Aware Projection (CLRP), a lightweight training-free framework. CLRP uses activation patching to identify the layer that causally mediates target-attribute disclosure, then applies a retain-aware projection that removes the target-attribute subspace while preserving same-identity evidence. Experiments across multiple widely used MLLMs with distinct architectures and parameter scales demonstrate the effectiveness of CLRP.

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Placed in MLLM jailbreaks and adversarial robustness, one of 16 directions in Multimodal LLMs / vision-language models. Builds on 19 of its 33 references that are in the map. First posted on arXiv 2 Aug 2026.

Builds on

19 of 33 references.

Closest ancestors · 5

Same direction, newest first

  1. 2026Visual-Noise Guided In-Context Distillation for Multimodal Large Language Model Unlearning26 May
  2. 2025Towards Benign Memory Forgetting for Selective Multimodal Large Language Model Unlearning25 Nov
  3. 2025MLLMEraser: Achieving Test-Time Unlearning in Multimodal Large Language Models through Activation Steering5 Oct
  4. 2025Unlearning Sensitive Information in Multimodal LLMs: Benchmark and Attack-Defense Evaluation1 May
  5. 2024Benchmarking Vision Language Model Unlearning via Fictitious Facial Identity Dataset5 Nov

Other references · 14

  1. 2022Locating and Editing Factual Associations in GPTcited by 2,609
  2. 2025Qwen-Image Technical Reportcited by 973
  3. 2023Steering Language Models With Activation Engineeringcited by 824
  4. 2024TOFU: A Task of Fictitious Unlearning for LLMscited by 507
  5. 2024Negative Preference Optimization: From Catastrophic Collapse to Effective Unlearningcited by 484
  6. 2020Null It Out: Guarding Protected Attributes by Iterative Nullspace Projectioncited by 442
  7. 2023LEACE: Perfect linear concept erasure in closed formcited by 254
  8. 2024Devils in Middle Layers of Large Vision-Language Models: Interpreting, Detecting and Mitigating Object Hallucinations via Attention Lenscited by 139
  9. 2024CLEAR: Character Unlearning in Textual and Visual Modalitiescited by 36
  10. 2024Targeted Unlearning with Single Layer Unlearning Gradientcited by 16
  11. 2026Multimodal Unlearning Across Vision, Language, Video, and Audio: Survey of Methods, Datasets, and Benchmarkscited by 6
  12. 2026SALMUBench: A Benchmark for Sensitive Association-Level Multimodal Unlearningcited by 3
2 more
  1. 2025Selective, Controlled and Domain-Agnostic Unlearning in Pretrained CLIP: A Training- and Data-Free Approachcited by 2
  2. 2025Erasing CLIP Memories: Non-Destructive, Data-Free Zero-Shot class Unlearning in CLIP Modelscited by 1

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2026Where Identity Lives: Localized, Retain-Free Identity Unlearning in Multimodal Large Language Models
  2. 2026AIM: Anchor Identity Features, Then Match for Multimodal Large Language Model Unlearning
  3. 2024Single Image Unlearning: Efficient Machine Unlearning in Multimodal Large Language Models
  4. 2025Towards Reasoning-Preserving Unlearning in Multimodal Large Language Models
  5. 2026Null Space Constrained Contrastive Visual Forgetting for MLLM Unlearning
  6. 2026ICU-Bench:Benchmarking Continual Unlearning in Multimodal Large Language Models
  7. 2025Not All Tokens Are Meant to Be Forgotten
  8. 2026Exploring and Bridging Knowledge Holes in Unlearned Multimodal Large Language Models