跳到正文
研踪PaperTrail
← 研踪 PaperTrail

arXiv:2610.32113 · v2

REACT: Representational Early Analysis and Control of Chain-of-Thought Trajectories

首次提交 2026-10-23 11:28
版本更新 2026-11-19 22:49
雷达观测 2026-11-20 14:52
arXiv ↗PDF ↗

摘要

Chain-of-thought (CoT) generation offers an opportunity to intervene before a reasoning model completes its answer. How does the safety effect of a single activation edit vary across early reasoning states, and what factors govern that variation? 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 DeepSeek-R1 distilled models, site-specific edits concentrate safety gains in the opening CoT tokens. Norm-matched controls on DeepSeek-R1-Distill-Llama-8B (8B) disentangle two factors behind this curve: holding an effective direction and edit budget fixed still yields much larger early (t1t_1-t5t_5) than later (t6t_6-t8t_8) gains, whereas the near-zero site-specific t3t_3 gain reflects higher optimisation variance despite rising risk-probe AUROC. Forced prefixes shift the high-gain region with the opening context, and a four-arm intervention shows that ~98% of the net gain at t2t_2 propagates through the changed next token. 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 https://anonymous.4open.science/r/REACT_official2-333B.

作者备注:33 pages, 5 figures

AI 论文解读

基于摘要 · 非全文审阅

此版本尚未生成 AI 解读。解读由后台任务生成并按论文版本保存,登录后共享已有结果。

版本差异

本站首次收录为 v2,暂无更早版本可对照。

观测到的版本

相邻记录可能跳过中间版本;首次发现 v2 及以上也会标为版本更新。

v2 · 2026-11-19 22:49 · 最新观测版本

REACT: Representational Early Analysis and Control of Chain-of-Thought Trajectories

Chain-of-thought (CoT) generation offers an opportunity to intervene before a reasoning model completes its answer. How does the safety effect of a single activation edit vary across early reasoning states, and what factors govern that variation? 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 DeepSeek-R1 distilled models, site-specific edits concentrate safety gains in the opening CoT tokens. Norm-matched controls on DeepSeek-R1-Distill-Llama-8B (8B) disentangle two factors behind this curve: holding an effective direction and edit budget fixed still yields much larger early (t1t_1-t5t_5) than later (t6t_6-t8t_8) gains, whereas the near-zero site-specific t3t_3 gain reflects higher optimisation variance despite rising risk-probe AUROC. Forced prefixes shift the high-gain region with the opening context, and a four-arm intervention shows that ~98% of the net gain at t2t_2 propagates through the changed next token. 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 https://anonymous.4open.science/r/REACT_official2-333B.

查看此版本解读 →查看此版本原文 ↗