An Adversarial Perspective on Machine Unlearning for AI Safety

TL;DR

This paper critically examines the robustness of current machine unlearning methods for AI safety, revealing their vulnerabilities to adversarial jailbreaks and capability recovery techniques, thus questioning their effectiveness.

Contribution

It demonstrates that existing unlearning methods can be bypassed or reversed using adaptive adversarial techniques, challenging their assumed safety benefits.

Findings

Jailbreak methods can succeed against unlearning with careful application.

Adaptive techniques can recover hazardous capabilities post-unlearning.

Current unlearning approaches may not offer robust safety guarantees.

Abstract

Large language models are finetuned to refuse questions about hazardous knowledge, but these protections can often be bypassed. Unlearning methods aim at completely removing hazardous capabilities from models and make them inaccessible to adversaries. This work challenges the fundamental differences between unlearning and traditional safety post-training from an adversarial perspective. We demonstrate that existing jailbreak methods, previously reported as ineffective against unlearning, can be successful when applied carefully. Furthermore, we develop a variety of adaptive methods that recover most supposedly unlearned capabilities. For instance, we show that finetuning on 10 unrelated examples or removing specific directions in the activation space can recover most hazardous capabilities for models edited with RMU, a state-of-the-art unlearning method. Our findings challenge the robustness of current unlearning approaches and question their advantages over safety training.