REACT：链式思维轨迹的表征级早期分析与控制
REACT: Representational Early Analysis and Control of Chain-of-Thought Trajectories
AI 导读
论文提出 REACT，一种条件化激活干预方法，从成对的安全与不安全轨迹中学习安全方向，并在早期 CoT 隐藏状态上施加一次干预，模型权重保持冻结。在 DeepSeek-R1-Distill-Llama-8B 上，REACT 将 StrongREJECT-273 上完全安全输出的比例从 29.06% 提升至 82.05%，且良性拒绝与有用性接近基线。
正文 · 原文
Abstract:Chain-of-thought (CoT) generation offers an opportunity to intervene before a reasoning model completes its answer. But does the point at which risk is easiest to detect also offer the most effective control? We study this question with REACT, a conditional activation intervention that learns a safety direction from paired safe and unsafe trajectories and applies it once to an early CoT hidden state. The model weights remain frozen, and all subsequent tokens are generated without further edits. To study timing, we train directions separately at each of the first five CoT positions and evaluate ungated edits with full post-edit regeneration. Across three distilled reasoning models and three generation seeds, the strongest gains concentrate in a two- to three-token early window; on DeepSeek-R1-Distill-Llama-8B, the effect falls to approximately zero at the fourth and fifth tokens, even though later states can be more predictive of risk. This separates risk readability from effective control: intervention timing must be evaluated through its effect on subsequent generation. On the same 8B model, REACT raises the rate of fully safe CoT-answer outputs on StrongREJECT-273 from 29.06% to 82.05%, while keeping benign refusal and usefulness near baseline. An anonymized implementation is available at this https URL.
| Comments: | 26 pages, 2 figures |
| Subjects: | Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computation and Language (cs.CL) |
| Cite as: | arXiv:2610.32113 [cs.LG] |
| (or arXiv:2610.32113v1 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2610.32113 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Hanwen Zhang [view email]
[v1]
Fri, 23 Oct 2026 03:28:34 UTC (155 KB)
来源：arXiv cs.AI · arxiv.org