NExT-Guard: Training-Free Streaming Safeguard without Token-Level Labels

TL;DR

NExT-Guard offers a training-free, real-time safety safeguard for large language models by leveraging latent features from pre-trained autoencoders, outperforming supervised methods without requiring token-level labels.

Contribution

It introduces a novel, training-free streaming safeguard framework that utilizes pretrained autoencoders to detect unsafe content in real-time without token-level supervision.

Findings

Outperforms supervised streaming safeguards in robustness

Works effectively across different models and risk scenarios

Enables low-cost, scalable deployment of real-time safety measures

Abstract

Large language models are increasingly deployed in streaming scenarios, rendering conventional post-hoc safeguards ineffective as they fail to interdict unsafe content in real-time. While streaming safeguards based on token-level supervised training could address this, they necessitate expensive annotations and suffer from severe overfitting. In this work, we challenge the paradigm that streaming safety must rely on token-level supervised training. Instead, it is an inherent capability of well-trained post-hoc safeguards, as they already encode token-level risk signals in hidden representations. Hence, we introduce NExT-Guard, a training-free framework that achieves streaming safeguards by monitoring interpretable latent features from Sparse Autoencoders (SAEs). It uses pretrained SAEs from publicly available base LLMs, enabling flexible, low-cost deployment without token-level supervision. Experimental results show that NExT-Guard outperforms both post-hoc and streaming safeguards based on supervised training, with superior robustness across models, SAE variants, and risk scenarios. These results make NExT-Guard a universal and scalable paradigm for real-time safety, accelerating the practical deployment of streaming safeguards.