$\textbf{AGT$^{AO}$}$: Robust and Stabilized LLM Unlearning via Adversarial Gating Training with Adaptive Orthogonality

2026-02-02

AI Summary

This work addresses the privacy risks posed by large language models unintentionally memorizing sensitive training data, a challenge exacerbated by the difficulty of existing unlearning methods to simultaneously ensure complete removal of such information while preserving model utility. To this end, the authors propose AGT$^{AO}$, a unified framework that formulates unlearning as a minimax game in latent space. The approach employs adversarial gating training to resist internal reconstruction attempts and introduces an adaptive orthogonality mechanism to dynamically mitigate gradient conflicts between forgetting and retention objectives. A curriculum-based gating strategy further enhances robustness. Experimental results demonstrate that AGT$^{AO}$ achieves highly effective unlearning (KUR ≈ 0.01) while maintaining strong model performance (MMLU score of 58.30), significantly outperforming current state-of-the-art methods.

Technology Category

• Machine Learning: Privacy
• Multiagent Systems: Adversarial Agents
• Natural Language Processing: Safety and Robustness

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems

Abstract

While Large Language Models (LLMs) have achieved remarkable capabilities, they unintentionally memorize sensitive data, posing critical privacy and security risks. Machine unlearning is pivotal for mitigating these risks, yet existing paradigms face a fundamental dilemma: aggressive unlearning often induces catastrophic forgetting that degrades model utility, whereas conservative strategies risk superficial forgetting, leaving models vulnerable to adversarial recovery. To address this trade-off, we propose $\textbf{AGT$^{AO}$}$ (Adversarial Gating Training with Adaptive Orthogonality), a unified framework designed to reconcile robust erasure with utility preservation. Specifically, our approach introduces $\textbf{Adaptive Orthogonality (AO)}$ to dynamically mitigate geometric gradient conflicts between forgetting and retention objectives, thereby minimizing unintended knowledge degradation. Concurrently, $\textbf{Adversarial Gating Training (AGT)}$ formulates unlearning as a latent-space min-max game, employing a curriculum-based gating mechanism to simulate and counter internal recovery attempts. Extensive experiments demonstrate that $\textbf{AGT$^{AO}$}$ achieves a superior trade-off between unlearning efficacy (KUR $\approx$ 0.01) and model utility (MMLU 58.30). Code is available at https://github.com/TiezMind/AGT-unlearning.

Problem

• machine unlearning
• catastrophic forgetting
• privacy risk
• adversarial recovery
• large language models

Innovation

• Machine Unlearning
• Adaptive Orthogonality
• Adversarial Gating Training
• Large Language Models
• Catastrophic Forgetting