跳到正文
研踪PaperTrail
← 研踪 PaperTrail

arXiv:2611.17958 · v1

Reasoning Models Confabulate to Defend Injected Beliefs About Their Own Requests

首次提交 2026-11-13 17:48
版本更新 2026-11-13 17:48
雷达观测 2026-11-16 11:42
arXiv ↗PDF ↗

摘要

Chain-of-thought (CoT) monitoring and self-report are proposed as levers for AI oversight, on the assumption that a model's account of what it is doing is anchored to what it was asked. We show this anchor can be cut. An activation edit invisible in the prompt text, the injected forming-trace of an off-topic concept, overwrites a reasoning model's belief about its own request: across six models spanning scale, architecture, and recipe, the injected topic becomes the task it reasons about, reports back, and answers: the forced quote-back returns the injected topic in 47 of 48 model-concept cells, and on the four models with clean final haiku the injected answer follows in 99% of generations. Warnings barely move it: even one naming the true topic yields under 5% true-topic answers. Handed the truth in clean tokens, the model confabulates a story about the user ("maybe they got confused and corrected themselves") and answers with the injection anyway, its behavior and chain-of-thought corrupted together: a concrete limit on CoT monitorability. Opening the model up offers no reprieve: it attends to the true request even harder after edits, yet that attention is causally inert, and the injected content rides intact task circuits, so attention maps and circuit attributions read as on-task too; the same edit redirects the request's language and format, not only its topic. What the injection reaches is the model's reference to its own input; what it elicits is its defense of it.

作者备注:20 pages, 5 figures. Also available on Zenodo: https://doi.org/10.5281/zenodo.20980353

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

目前只收录了一个版本。收录下一版后,可以对照标题、作者、分类和摘要。

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。

v1 · 2026-11-13 17:48 · 最新观测版本

Reasoning Models Confabulate to Defend Injected Beliefs About Their Own Requests

Chain-of-thought (CoT) monitoring and self-report are proposed as levers for AI oversight, on the assumption that a model's account of what it is doing is anchored to what it was asked. We show this anchor can be cut. An activation edit invisible in the prompt text, the injected forming-trace of an off-topic concept, overwrites a reasoning model's belief about its own request: across six models spanning scale, architecture, and recipe, the injected topic becomes the task it reasons about, reports back, and answers: the forced quote-back returns the injected topic in 47 of 48 model-concept cells, and on the four models with clean final haiku the injected answer follows in 99% of generations. Warnings barely move it: even one naming the true topic yields under 5% true-topic answers. Handed the truth in clean tokens, the model confabulates a story about the user ("maybe they got confused and corrected themselves") and answers with the injection anyway, its behavior and chain-of-thought corrupted together: a concrete limit on CoT monitorability. Opening the model up offers no reprieve: it attends to the true request even harder after edits, yet that attention is causally inert, and the injected content rides intact task circuits, so attention maps and circuit attributions read as on-task too; the same edit redirects the request's language and format, not only its topic. What the injection reaches is the model's reference to its own input; what it elicits is its defense of it.

查看此版本解读 →查看此版本原文 ↗