arXiv:2601.04394cs.CL2026-01Conference of the …

用外部网络动态修正大模型幻觉与安全问题,无需微调参数。

ARREST: Adversarial Resilient Regulation Enhancing Safety and Truth in Large Language Models

  • 通过检测隐空间漂移,外部网络实时干预纠正错误输出。
  • 在不微调模型的前提下,同时提升事实性与安全性表现。
  • 适合需要高安全性和可靠性的生成场景,如医疗、法律应用。

人类认知依赖复杂的神经化学过程,在想象与现实间波动,并能自我纠正因细微偏差导致的幻觉或不安全联想。近年来,大语言模型在多种任务中表现出色,但仍缺乏人类认知中对真实性和安全性的平衡能力。我们认为,事实性与安全性失败均源于模型隐空间表征的错位,而非独立的对齐问题。我们提出一种假设:外部网络可学习识别激活波动,从而在不微调模型参数的情况下,选择性干预以将错误内容修正为真实、安全输出。基于此,我们构建了ARREST(Adversarial Resilient Regulation Enhancing Safety and Truth)框架,可识别并纠正漂移特征,支持软拒绝、硬拒绝及事实修正。实验表明,该方法不仅能有效调节表征错位,且在生成软拒绝方面较强化学习人类反馈(RLHF)对齐模型更具灵活性,得益于对抗训练机制。代码已开源。

原文摘要 · Abstract (English)

Human cognition, driven by complex neurochemical processes, oscillates between imagination and reality and learns to self-correct whenever such subtle drifts lead to hallucinations or unsafe associations. In recent years, LLMs have demonstrated remarkable performance in a wide range of tasks. However, they still lack human cognition to balance factuality and safety. Bearing the resemblance, we argue that both factual and safety failures in LLMs arise from a representational misalignment in their latent activation space, rather than addressing those as entirely separate alignment issues. We hypothesize that an external network, trained to understand the fluctuations, can selectively intervene in the model to regulate falsehood into truthfulness and unsafe output into safe output without fine-tuning the model parameters themselves. Reflecting the hypothesis, we propose ARREST (Adversarial Resilient Regulation Enhancing Safety and Truth), a unified framework that identifies and corrects drifted features, engaging both soft and hard refusals in addition to factual corrections. Our empirical results show that ARREST not only regulates misalignment but is also more versatile compared to the RLHF-aligned models in generating soft refusals due to adversarial training. We make our codebase available at https://github.com/sharanya-dasgupta001/ARREST.

大模型安全幻觉纠正对抗训练

Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品;中文卡片由大模型生成,请以原文为准。