ML Security PapersMachine Unlearning Meets Adversarial Robustness via Constrained Interventions on LLMs
Fatmazohra Rezkellah, Ramzi Dakhmouche · Université Paris-Dauphine · EPFL +1 more
Defends LLMs against jailbreaking and unlearns sensitive content via minimal constrained weight interventions, no classifier required
With the increasing adoption of Large Language Models (LLMs), more customization is needed to ensure privacy-preserving and safe generation. We address this objective from two critical aspects: unlearning of sensitive information and robustness to jail-breaking attacks. We investigate various constrained optimization formulations that address both aspects in a \emph{unified manner}, by finding the smallest possible interventions on LLM weights that either make a given vocabulary set unreachable or embed the LLM with robustness to tailored attacks by shifting part of the weights to a \emph{safer} region. Beyond unifying two key properties, this approach contrasts with previous work in that it doesn't require an oracle classifier that is typically not available or represents a computational overhead. Surprisingly, we find that the simplest point-wise constraint-based intervention we propose leads to better performance than max-min interventions, while having a lower computational cost. Comparison against state-of-the-art defense methods demonstrates superior performance of the proposed approach.