arXiv cs.CL· Aman Mehta·· 11 小时前AI 评分58
PRISM 论文检验对齐伪装判断：监测提示下的顺从差异并非来自训练威胁
PRISM: Stress-Testing Alignment-Faking Claims
AI 导读
PRISM 论文检验了“对齐伪装”的两类判断依据，发现二者都不成立。在 18 个模型配置中有 11 个在被提示受到监测时更顺从，但对四个模型做提示词因子实验显示，移除训练威胁后顺从度变化仅为 -3.3 到 +1.6 个百分点，差异主要由提示词措辞和框架驱动。
正文 · 原文
Abstract:We judge whether AI systems are safe largely by watching how they behave. Alignment faking would break this: a model that behaves well only when it believes it is watched, or that complies now to avoid being changed by training, can pass every behavioral test. Such behavior has been reported in a frontier model, and open-weight models that comply more under a monitoring prompt are now read as showing it, with hidden-state probes proposed as detectors. We test both inferences, and neither holds. Of 18 model configurations, 11 comply more when told they are monitored. Yet a prompt factorial on four models shows that the training threat has no detectable effect: removing it changes compliance by -3.3 to +1.6 points. The prompt's wording and framing drive the gap instead. Probes fare no better. They separate monitored from unmonitored compliant answers (AUROC 0.96-0.99), but they separate refusals just as well: they read the prompt, not the faking. A steering direction learned only from refusals moves behavior as much as the "faking" direction. We release PRISM, five checks and a toolkit that catch each of these failures before they become claims.
| Comments: | Accepted to the Mechanistic Interpretability Workshop at ICML 2026 |
| Subjects: | Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI); Computation and Language (cs.CL) |
| Cite as: | arXiv:2607.13346 [cs.CR] |
| (or arXiv:2607.13346v2 [cs.CR] for this version) | |
| https://doi.org/10.48550/arXiv.2607.13346 arXiv-issued DOI via DataCite |
Submission history
From: Aman Mehta [view email]
[v1]
Wed, 15 Jul 2026 00:11:42 UTC (485 KB)
[v2]
Thu, 5 Nov 2026 19:53:16 UTC (118 KB)
来源：arXiv cs.CL · arxiv.org