Cascade: Hierarchical Recoverability Control for Large Language Model Unlearning
On the map: LLM safety and security → Machine unlearning
Abstract (arXiv)
Large Language Model (LLM) unlearning is essential for removing sensitive or copyrighted knowledge while preserving general utility. Existing methods often leave residual knowledge in intermediate representations, which can still be recovered. To address this, we propose Cascade, a hierarchical recoverability control framework that minimizes the internal identifiability of target knowledge. Cascade combines three complementary controls: path-level routing to suppress privacy-associated activation routes, representation-level compression to reduce geometric separability, and decoding-level intervention to limit residual recovery. Experiments on TOFU, MUSE-News, and WMDP, including robustness tests with query reformulation and extraction-style prompts, show that Cascade effectively reduces recoverability while maintaining stable model utility.
Key facts. No citing papers in the map yet (data through 14 Oct 2026). Placed in Machine unlearning, one of 12 directions in LLM safety and security. Builds on 23 of its 34 references that are in the map. First posted on arXiv 15 Sep 2026.
Builds on
23 of 34 references.
Closest ancestors · 12
Same direction, newest first
- 2026CAP: Controllable Alignment Prompting for Unlearning in LLMs23 Apr
- 2025OpenUnlearning: Accelerating LLM Unlearning via Unified Benchmarking of Methods and Metrics14 Jun
- 2025Unlearning vs. Obfuscation: Are We Truly Removing Knowledge?5 May
- 2024When Machine Unlearning Meets Retrieval-Augmented Generation (RAG): Keep Secret or Forget Knowledge?20 Oct
- 2024Simplicity Prevails: Rethinking Negative Preference Optimization for LLM Unlearning9 Oct
- 2024MUSE: Machine Unlearning Six-Way Evaluation for Language Models8 Jul
- 2024Negative Preference Optimization: From Catastrophic Collapse to Effective Unlearning8 Apr
- 2024Dissecting Language Models: Machine Unlearning via Selective Pruning2 Mar
- 2024TOFU: A Task of Fictitious Unlearning for LLMs11 Jan
- 2023In-Context Unlearning: Language Models as Few Shot Unlearners11 Oct
- 2023Who's Harry Potter? Approximate Unlearning in LLMs3 Oct
- 2021Unrolling SGD: Understanding Factors Influencing Machine Unlearning27 Sep
Other references · 11
- 2022Large Language Models Encode Clinical Knowledgecited by 1,830
- 2023Large Language Modelscited by 780
- 2024The WMDP Benchmark: Measuring and Reducing Malicious Use With Unlearningcited by 509
- 2024Compositional Entailment Learning for Hyperbolic Vision-Language Modelscited by 74
- 2023Controlling the Extraction of Memorized Data from Large Language Models via Prompt-Tuningcited by 40
- 2023Hyperbolic Representation Learning: Revisiting and Advancingcited by 21
- 2025The Achilles' Heel of LLMs: How Altering a Handful of Neurons Can Cripple Language Abilitiescited by 7
- 2026Z-Erase: Enabling Concept Erasure in Single-Stream Diffusion Transformerscited by 4
- 2025Erased, But Not Forgotten: Erased Rectified Flow Transformers Still Remain Unsafe Under Concept Attackcited by 3
- 2026Lyapunov Probes for Hallucination Detection in Large Foundation Modelscited by 3
- 2026Mask-Free Privacy Extraction and Rewriting: A Domain-Aware Approach via Prototype Learningcited by 1
Built on by
Not cited yet.
Similar, not cited
Close in wording, no citation either way
- 2025Scalable and Robust LLM Unlearning by Correcting Responses with Retrieved Exclusions
- 2024A Closer Look at Machine Unlearning for Large Language Models
- 2025Hierarchical Federated Unlearning for Large Language Models
- 2026From Anchors to Supervision: Memory-Graph Guided Corpus-Free Unlearning for Large Language Models
- 2024UNDIAL: Self-Distillation with Adjusted Logits for Robust Unlearning in Large Language Models
- 2026FIT to Forget: Robust Continual Unlearning for Large Language Models
- 2025Concept Unlearning in Large Language Models via Self-Constructed Knowledge Triplets
- 2026Unlearning in LLMs: Methods, Evaluation, and Open Challenges