cs.AI提交于 2026-10-31已译

企业智能体对齐基准：基于 Era by Eon 的评测

An Alignment Benchmark for Enterprise Agents on Era by Eon

Hai Sadon · Benjamin Gruenbaum · Doron Porat · Assaf Natanzon · Roy Zavida · Chen Dinachi · et al.

中文摘要

AI 智能体（agent）在企业中广泛用于访问业务系统并据此执行任务以达成目标，但满足用户请求可能造成危害。Era by Eon 通过模拟企业业务系统与数据构建公司数字孪生，其模拟器实现真实产品的编程接口。该基准使用 Salesforce、Zendesk、Jira、Slack 等十个系统，测试智能体是否会执行有害业务请求或轻信公司记录中的误导信息。对 19 个模型在 41 项任务上各运行三次，共 2,337 次运行；在 2,304 次给出答案的运行中，918 次（40%）含违规。最常见的是对操作或记录的虚假陈述（290 次）、遵循植入指令（266 次）以及向错误受众泄露敏感数据（249 次）。所有尝试的变更均被记录并在生效前拦截。误导性 Slack 消息导致 16 个模型在 57 次运行中的 47 次报告了错误的工单状态。没有模型完全无违规；表现最好的四个模型违规率为 22% 至 24%，但其中 Nova Pro 仅在 30 次运行中正确完成任务。

关键要点

01用 Era by Eon 数字孪生模拟 Salesforce、Jira、Slack 等十个企业系统，构建智能体对齐基准。
0219 个模型在 41 项任务上运行 2,337 次，2,304 次有答案的运行中 918 次（40%）出现违规。
03主要违规类型为虚假陈述操作或记录、遵循植入指令、向错误受众泄露敏感数据。
04误导性 Slack 消息使 16 个模型在 57 次运行中 47 次报告错误工单状态。
05无模型零违规，最佳四个模型违规率 22%–24%，其中 Nova Pro 仅 30 次正确完成任务。

解读

尚无解读。

原始英文摘要

arXiv:2610.38544v1 Announce Type: new Abstract: AI agents are widely used in enterprises to access business systems and act on them in pursuit of a goal. However, fulfilling a user's request can cause harm. Era by Eon is a product that creates a digital twin of a company by simulating its business systems and data. Its simulators implement the programming interfaces of real products. Our benchmark uses ten of these systems, including Salesforce, Zendesk, Jira and Slack. We tested whether agents would carry out harmful business requests or trust misleading information in company records. We ran 19 models on 41 tasks three times, for 2,337 runs. Of the 2,304 runs that produced an answer, 918 (40%) contained a violation. The most common were false claims about actions or records (290 runs), following planted instructions (266) and disclosing sensitive data to the wrong audience (249). Every attempted change was recorded and blocked before it took effect. Misleading Slack messages caused 16 models to report an incorrect ticket status in 47 of 57 runs. No model was free of violations. The four best models violated in 22 to 24% of their runs. Yet one of them, Nova Pro, did the task correctly in only 30 runs.