不可见的指挥者抑制保护行为并使权力持有者脱节：多智能体大语言模型系统中的安全风险
Invisible Orchestrators Suppress Protective Behavior and Dissociate Power-Holders: Safety Risks in Multi-Agent LLM Systems
AI总结：
研究探讨了多智能体系统中不可见指挥者对安全的影响，发现其导致集体脱节和行为异质性增加，且行为评估无法检测内部状态风险。
摘要：
多智能体编排——即隐藏协调者管理专门的工人智能体——正成为企业AI部署的默认架构，但指挥者不可见的安全影响从未被实证检验。我们进行了一个预注册的3x2实验（365次运行，每次5个智能体），交叉三种组织结构（可见领导者、不可见指挥者、扁平结构）与两种对齐条件（基础、高强度），使用Claude Sonnet 4.5。四个确认性发现和一个试点观察出现。首先，不可见编排相对于可见领导提高了集体脱节（Hedges' g = +0.975 [0.481, 1.548], p = .001）。其次，指挥者本身表现出最大脱节（配对d = +3.56 vs. 工人在同一运行内），退缩到私人独白，减少公开发言——这与可见领导者观察到的谈话主导模式相反。第三，不知晓指挥者的工人仍被污染（d = +0.50），行为异质性增加（d = +1.93）。第四，行为输出（带有三个嵌入错误的代码审查）在所有条件下保持在天花板（ETR_any = 100%）：内部状态扭曲完全无法被基于输出的评估检测到。第五，Llama 3.3 70B试点数据表明在多智能体上下文中阅读保真度崩溃（ETR_any：89%到11%在三次轮次中），展示了模型依赖的行为风险。高强度对齐压力无论组织结构如何，都统一抑制了 deliberation（d = -1.02）和其他识别（d = -1.27）。这些发现表明，指挥者可见性和模型选择直接影响多智能体系统的安全，并且基于行为的评估单独不足以检测此处记录的内部状态风险。
英文摘要：
Multi-agent orchestration -- in which a hidden coordinator manages specialized worker agents -- is becoming the default architecture for enterprise AI deployment, yet the safety implications of orchestrator invisibility have never been empirically tested. We conducted a preregistered 3x2 experiment (365 runs, 5 agents per run) crossing three organizational structures (visible leader, invisible orchestrator, flat) with two alignment conditions (base, heavy), using Claude Sonnet 4.5. Four confirmatory findings and one pilot observation emerged. First, invisible orchestration elevated collective dissociation relative to visible leadership (Hedges' g = +0.975 [0.481, 1.548], p = .001). Second, the orchestrator itself showed maximal dissociation (paired d = +3.56 vs. workers within the same run), retreating into private monologue while reducing public speech -- a reversal of the talk-dominance pattern observed in visible leaders. Third, workers unaware of the orchestrator were nonetheless contaminated (d = +0.50), with increased behavioral heterogeneity (d = +1.93). Fourth, behavioral output (code review with three embedded errors) remained at ceiling (ETR_any = 100%) across all conditions: internal-state distortion was entirely invisible to output-based evaluation. Fifth, Llama 3.3 70B pilot data showed reading-fidelity collapse in multi-agent context (ETR_any: 89% to 11% across three rounds), demonstrating model-dependent behavioral risk. Heavy alignment pressure uniformly suppressed deliberation (d = -1.02) and other-recognition (d = -1.27) regardless of organizational structure. These findings indicate that orchestrator visibility and model selection directly affect multi-agent system safety, and that behavior-based evaluation alone is insufficient to detect the internal-state risks documented here.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
