ChatGPT Workspace Agents for Enterprise and Business

ChatGPT Workspace Agents for Enterprise and Business

Summary (EN)

OpenAI has documented a new Workspace Agents capability for ChatGPT Business and Enterprise, describing a system for creating, sharing, and operating organization-level agents across ChatGPT and Slack. The article focuses heavily on operational controls rather than a marketing narrative, which is itself informative. Workspace Agents can be connected to Slack channels and can use shared app connections, but OpenAI repeatedly warns that poor credential design can allow people invoking an agent to access data or perform actions through the publisher’s permissions. The documentation recommends service-account based shared authentication, least-privilege scoping, trusted Slack distribution, and explicit write-action safety for workflows with elevated risk. It also highlights role-based access controls, cautioning that allowing publishing with personal connections can create unintended data exposure or action delegation. In effect, OpenAI is pushing agent deployment beyond individual productivity into managed organizational automation, but is doing so with visible concern about data exfiltration, overbroad permissions, and unsafe write actions. That makes the release notable not just as a product feature but as an enterprise design signal. The company is treating agent sharing, connector identity, and action authorization as first-order product surfaces, not afterthoughts. For senior editors tracking practical AI adoption, this matters because it shows where commercialization is moving: enterprises want reusable agents embedded in existing collaboration systems, yet the gating issue is increasingly governance, credential architecture, and permission containment rather than raw model intelligence alone.

Summary (ZH)

OpenAI 已为 ChatGPT Business 和 Enterprise 发布 Workspace Agents 相关文档，描述了一套可在组织层面创建、共享并在 ChatGPT 与 Slack 中运行代理系统的能力。值得注意的是，这篇文档的重点并不是产品宣传，而是部署与治理控制本身，这一点本身就具有信息价值。Workspace Agents 可以连接到 Slack 频道，也可以使用共享的应用连接，但 OpenAI 反复提醒，如果凭证设计不当，调用代理的用户可能会通过发布者的权限访问数据或执行操作。官方建议使用基于服务账号的共享认证、最小权限配置、受信任的 Slack 分发范围，以及针对高风险流程启用明确的写操作安全机制。文档还特别强调基于角色的访问控制，警告说如果允许用户用个人连接发布代理，可能会带来意外的数据暴露或操作代理风险。换句话说，OpenAI 正把代理从个人效率工具推进到可管理的组织自动化层，但同时公开承认其核心难题在于数据外泄、权限过宽以及危险写操作的治理。这使得此次更新的重要性不仅在于“又一个代理功能”，更在于它释放出的企业产品信号，即代理共享、连接器身份和动作授权已经成为一等公民级别的产品表面，而不再只是附属设置。对于关注 AI 落地的编辑而言，这说明企业真正需要的是能嵌入现有协作系统的可复用代理，而决定其能否落地的关键，越来越是治理架构与权限边界，而不是模型本身是否再强一点。

Source

https://help.openai.com/en/articles/20001143-chatgpt-workspace-agents-for-enterprise-and-business