Large Language Lobotomy: Jailbreaking Mixture-of-Experts via Expert Silencing

Authors: Jona te Lintelo, Lichao Wu, Stjepan Picek
Published: 2026-02-09

Source: https://arxiv.org/abs/2602.08741

AIにより推定されたラベル

※ こちらのラベルはAIによって自動的に追加されました。そのため、正確でないことがあります。
詳細は文献データベースについてをご覧ください。

Abstract

The rapid adoption of Mixture-of-Experts (MoE) architectures marks a major shift in the deployment of Large Language Models (LLMs). MoE LLMs improve scaling efficiency by activating only a small subset of parameters per token, but their routing structure introduces new safety attack surfaces. We find that safety-critical behaviors in MoE LLMs (e.g., refusal) are concentrated in a small set of experts rather than being uniformly distributed. Building on this, we propose Large Language Lobotomy (L³), a training-free, architecture-agnostic attack that compromises safety alignment by exploiting expert routing dynamics. L³ learns routing patterns that correlate with refusal, attributes safety behavior to specific experts, and adaptively silences the most safety-relevant experts until harmful outputs are produced. We evaluate L³ on eight state-of-the-art open-source MoE LLMs and show that our adaptive expert silencing increases average attack success from 7.3