Do Prompts Guarantee Safety? Mitigating Toxicity from LLM Generations through Subspace Intervention

TL;DR

This paper introduces a subspace intervention method to effectively reduce toxicity in large language models' outputs, maintaining fluency and safety without significant computational costs.

Contribution

The paper presents a novel subspace intervention technique that identifies and suppresses toxic patterns in LLM representations, improving safety while preserving generation quality.

Findings

Reduces toxicity by 8-20% across multiple LLMs.

Maintains comparable fluency and coherence in generated content.

Achieves strong mitigation performance with minimal inference impact.

Abstract

Large Language Models (LLMs) are powerful text generators, yet they can produce toxic or harmful content even when given seemingly harmless prompts. This presents a serious safety challenge and can cause real-world harm. Toxicity is often subtle and context-dependent, making it difficult to detect at the token level or through coarse sentence-level signals. Moreover, efforts to mitigate toxicity often face a trade-off between safety and the coherence, or fluency of the generated text. In this work, we present a targeted subspace intervention strategy for identifying and suppressing hidden toxic patterns from underlying model representations, while preserving overall ability to generate safe fluent content. On the RealToxicityPrompts, our method achieves strong mitigation performance compared to existing baselines, with minimal impact on inference complexity. Across multiple LLMs, our approach reduces toxicity of state-of-the-art detoxification systems by 8-20%, while maintaining comparable fluency. Through extensive quantitative and qualitative analyses, we show that our approach achieves effective toxicity reduction without impairing generative performance, consistently outperforming existing baselines.