Alignment-Enhanced Decoding:Defending via Token-Level Adaptive Refining of Probability Distributions

TL;DR

This paper introduces Alignment-Enhanced Decoding (AED), a novel adaptive decoding method that improves language model safety by addressing alignment failures at the token level through a feedback-driven refinement process.

Contribution

The paper proposes a new decoding technique that adaptively combines logits to enhance safety and alignment in language models, focusing on root causes of jailbreak vulnerabilities.

Findings

Effective in reducing harmful outputs across multiple models

Maintains helpfulness while improving safety

Validated on various jailbreak scenarios

Abstract

Large language models are susceptible to jailbreak attacks, which can result in the generation of harmful content. While prior defenses mitigate these risks by perturbing or inspecting inputs, they ignore competing objectives, the underlying cause of alignment failures. In this paper, we propose Alignment-Enhanced Decoding (AED), a novel defense that employs adaptive decoding to address the root causes of jailbreak issues. We first define the Competitive Index to quantify alignment failures and utilize feedback from self-evaluation to compute post-alignment logits. Then, AED adaptively combines AED and post-alignment logits with the original logits to obtain harmless and helpful distributions. Consequently, our method enhances safety alignment while maintaining helpfulness. We conduct experiments across five models and four common jailbreaks, with the results validating the effectiveness of our approach. Code is available at https://github.com/GIGABaozi/AED.git.