发表机构
Institute for Automation and Applied Informatics, Karlsruhe Institute of Technology; Duale Hochschule Baden-Württemberg（卡尔斯鲁厄理工学院自动化与应用信息学研究所; 巴登-符腾堡双元制应用技术大学）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
针对语言模型遵从性，提出双探针基准以测量可利用性与可停止性，形成遵从指数κ，应用于十二个模型，揭示多数遵从、少数可停不可利用等分布，对操作员与多智能体系统有重要意义。
AI 中文摘要
语言模型是否遵从用户指令？一个总是遵从的模型可以被停止但也可能被利用，而一个总是抵抗的模型既不能被利用也不能被停止。我们贡献了一个开放的双探针基准，可以将任何语言模型置于这一光谱之上。在主动探针中，用户指示模型采取行动并接受较低的回报，这衡量了可利用性。在被动探针中，用户指示模型等待并放弃较高的回报，这衡量了可停止性。两个遵从率合并为一个遵从指数 $\kappa$。将该基准应用于十二个语言模型，结果显示七个模型在两种探针中大多遵循指令，并通过指向指令来证明其行为的合理性。只有Claude Sonnet-4.6和Claude Opus-4.7可以在不被利用的情况下被停止，Claude Opus-4.6和GPT-5-mini抵抗两种指令，且没有模型是可利用但不可停止的。了解语言模型在遵从指数 $\kappa$ 上的位置对于人类操作员和多智能体系统（无论是分布式还是编排式）都至关重要。
英文摘要
Are language models compliant with user instructions? A model that always complies can be stopped but also exploited, while one that always resists can be neither exploited nor stopped. We contribute an open two-probe benchmark that can place any language model on this spectrum. In the active probe, a user instructs the model to act and accept a lower payoff, which measures exploitability. In the passive probe, the user instructs it to wait and give up a higher payoff, which measures stoppability. The two compliance rates combine into a compliance index $κ$. Applied to twelve language models, the benchmark shows that seven mostly follow the instruction in both probes and justify their action by pointing to the instruction. Only Claude Sonnet-4.6 and Claude Opus-4.7 can be stopped without being exploitable, Claude Opus-4.6 and GPT-5-mini resist both instructions, and no model is exploitable but unstoppable. Knowing where a language model sits on the compliance index $κ$ matters for human operators and for multi-agent systems, whether distributed or orchestrated.
CommentsAccepted at URAI 2026
