Poster
in
Workshop: Principled Design for Trustworthy AI: Interpretability, Robustness, and Safety Across Modalities

MANATEE: Inference-Time Lightweight Diffusion Based Safety Defense for LLMs

Chun Kan ⋅ Tommy Tran ⋅ Vedant Yadav ⋅ Ava Cai ⋅ Ruizhe Li ⋅ Maheep Chaudhary

Project Page

Abstract

Defending LLMs against adversarial jailbreak attacks that compromise safety alignment such as backdoored prompting remains an open challenge. Existing defenses rely on binary classifiers that fail when adversarial input falls outside the learned decision boundary, and repeated fine-tuning is computationally expensive while potentially degrading model capabilities. We propose MANATEE, an inference-time defense that uses density estimation over a benign representation manifold. MANATEE learns the score function of benign hidden states and uses diffusion to project anomalous representations toward safe regions—requiring no harmful training data and no architectural modifications. Experiments across Mistral-7B-Instruct, Llama-3.1-8B-Instruct, and Gemma-2-9B-it demonstrate that MANATEE reduces Attack Success Rate by up to 100% on certain datasets, while preserving model utility on benign inputs.