mlsanityData to 7 Oct 2026

SPACE: Source-free Proxy Anchor Concept Erasure for MLLMs

arXiv 2606.09868 ↗First posted 1 Jun 2026cs.LG cs.AIcited by 0

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

also in 1 more
Abstract (arXiv)

As Multimodal Large Language Models (MLLMs) face growing privacy risks and regulatory constraints, machine unlearning (MU) has emerged as a crucial solution for removing sensitive data while preserving model performance. However, existing MU methods typically rely on visual data of the target concepts, which is often unavailable due to strict data retention policies, thus creating a demand for source-free unlearning approaches that operate without access to the target data. In this work, we propose Source-free Proxy Anchor Concept Erasure (SPACE), the first source-free unlearning framework specialized for MLLMs. SPACE consists of two stages: (1) Text-Guided Proxy Anchor Selection (TPAS), which retrieves semantically aligned proxy anchors from the shared feature space. (2) Dual-Constraint Semantic Isolation (DCSI), which optimizes these anchors to indirectly erase target concepts. DCSI confines updates to the null space of retained knowledge, ensuring structural integrity. We theoretically prove that SPACE strictly bounds the perturbation on retained knowledge and maximizes feature spectral entropy, thereby maintaining the model's performance. Furthermore, extensive experiments across six datasets show that SPACE achieves performance comparable to that of state-of-the-art data-dependent methods, validating its effectiveness in source-free MU scenarios. The source code will be released.

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 37 of its 44 references that are in the map. First posted on arXiv 1 Jun 2026.

Builds on

37 of 44 references.

Closest ancestors · 7

Same direction, newest first

  1. 2025Cross-Modal Unlearning via Influential Neuron Path Editing in Multimodal Large Language Models10 Nov
  2. 2025OFFSIDE: Benchmarking Unlearning Misinformation in Multimodal Large Language Models26 Oct
  3. 2025PEBench: A Fictitious Dataset to Benchmark Machine Unlearning for Multimodal Large Language Models16 Mar
  4. 2025Modality-Aware Neuron Pruning for Unlearning in Multimodal Large Language Models21 Feb
  5. 2025MMUnlearner: Reformulating Multimodal Machine Unlearning in the Era of Multimodal Large Language Models16 Feb
  6. 2024Protecting Privacy in Multimodal Large Language Models with MLLMU-Bench29 Oct
  7. 2024Single Image Unlearning: Efficient Machine Unlearning in Multimodal Large Language Models21 May

Other references · 30

  1. 2015Deep Residual Learning for Image Recognitioncited by 51,809
  2. 2021Learning Transferable Visual Models From Natural Language Supervisioncited by 33,456
  3. 2020An Image is Worth 16x16 Words: Transformers for Image Recognition at Scalecited by 29,915
  4. 2023Visual Instruction Tuningcited by 7,792
  5. 2023InternVL: Scaling up Vision Foundation Models and Aligning for Generic Visual-Linguistic Taskscited by 2,534
  6. 2019Towards VQA Models That Can Readcited by 1,696
  7. 2024TOFU: A Task of Fictitious Unlearning for LLMscited by 507
  8. 2024Negative Preference Optimization: From Catastrophic Collapse to Effective Unlearningcited by 484
  9. 2022Knowledge Unlearning for Mitigating Privacy Risks in Language Modelscited by 424
  10. 2023Unified Concept Editing in Diffusion Modelscited by 377
  11. 2022Visual Classification via Description from Large Language Modelscited by 331
  12. 2023Large Language Model Unlearningcited by 288
18 more
  1. 2022What does a platypus look like? Generating customized prompts for zero-shot image classificationcited by 263
  2. 2021Fast Yet Effective Machine Unlearningcited by 209
  3. 2022Zero-Shot Machine Unlearningcited by 125
  4. 2024Machine Unlearning of Pre-trained Large Language Modelscited by 118
  5. 2024Large Language Model Unlearning via Embedding-Corrupted Promptscited by 112
  6. 2024EraseAnything: Enabling Concept Erasure in Rectified Flow Transformerscited by 93
  7. 2025Unlearning Isn't Deletion: Investigating Reversibility of Machine Unlearning in LLMscited by 41
  8. 2025Interpreting the linear structure of vision-language model embedding spacescited by 32
  9. 2024Human-like object concept representations emerge naturally in multimodal large language modelscited by 24
  10. 2025ReLearn: Unlearning via Learning for Large Language Modelscited by 20
  11. 2026Mass Concept Erasure in Diffusion Models with Concept Hierarchycited by 14
  12. 2024Textual Unlearning Gives a False Sense of Unlearningcited by 13
  13. 2025LoTUS: Large-Scale Machine Unlearning with a Taste of Uncertaintycited by 12
  14. 2024Towards Natural Machine Unlearningcited by 11
  15. 2025Towards Source-Free Machine Unlearningcited by 6
  16. 2024Toward Efficient Data-Free Unlearningcited by 4
  17. 2024CLIP Adaptation by Intra-modal Overlap Reductioncited by 2
  18. 2025Zero-Shot Machine Unlearning with Proxy Adversarial Data Generationcited by 1

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2026Open-MMUnlearning: Unifying Methods and Evaluation for MLLM Unlearning
  2. 2026Exploring and Bridging Knowledge Holes in Unlearned Multimodal Large Language Models
  3. 2026MeGU: Machine-Guided Unlearning with Target Feature Disentanglement
  4. 2025Align-then-Unlearn: Embedding Alignment for LLM Unlearning
  5. 2025AUVIC: Adversarial Unlearning of Visual Concepts for Multi-modal Large Language Models
  6. 2026MLUBench: A Benchmark for Lifelong Unlearning Evaluation in MLLMs
  7. 2025SafeEraser: Enhancing Safety in Multimodal Large Language Models through Multimodal Machine Unlearning
  8. 2026Toward Fine-Grained Forgetting:Attribute Unlearning for Multimodal Large Language Models