Making LLMs Truly Forget: Deep Unlearning by Searching, Selecting, and Severing Knowledge Paths

2026-09-28

AI Summary

Existing unlearning techniques for large language models merely remove surface-level facts, allowing target knowledge to be recovered through multi-hop reasoning. This work proposes a deep unlearning framework that, for the first time, disrupts relational structures. Specifically, it constructs support subgraphs via model-specific knowledge graph extraction and internal representation probing, introduces a confidence-aware filtering mechanism, and applies graph-theoretic minimum cut algorithms to precisely sever all potential recovery paths. Experimental results demonstrate that this approach significantly deepens the unlearning effect and effectively blocks multi-hop reasoning vulnerabilities while preserving the general capabilities of the model.

Abstract

While an unlearned language model may no longer recall a fact directly, the fact often remains recoverable through multi-hop reasoning over related knowledge. Most existing unlearning techniques overlook this vulnerability, targeting facts in isolation while leaving their supporting knowledge intact. To achieve true forgetting, we propose a general deep unlearning framework compatible with existing unlearning algorithms. Our approach adaptively explores both explicit responses and latent internal representations to discover valid reasoning paths, compiles them into a confidence-aware supporting subgraph, and we apply a graph minimum cut to sever all recovery paths while preserving unrelated knowledge. To rigorously evaluate deep unlearning, we introduce a model-specific pipeline that extracts and completes knowledge graphs from raw text, filtering them by calibrated model confidence to reflect what the model genuinely retains. Comprehensive experiments demonstrate that selectively unlearning supporting knowledge yields substantially deeper forgetting than superficial methods while preserving model utility, highlighting that genuine unlearning requires breaking the relational structures that enable factual reconstruction.

Problem

• Deep Unlearning
• Large Language Models
• Multi-hop Reasoning
• Knowledge Recovery
• Machine Unlearning

Innovation

• Deep Unlearning
• Multi-hop Reasoning
• Graph Minimum Cut
• Knowledge Graph
• Large Language Models