ServiceNow AI Passport Sets Limits on Agent Actions

ServiceNow says its AI Passport will let companies define which systems an AI agent can access and what actions it can take, including in systems from other providers. The company demonstrated the digital credential at a press briefing on October 15.

In the demonstration, an agent could issue refunds below $25 without human review. It processed an $18 refund, while a $40 request was stopped for human review. The agent was a ready-made third-party product that could also track orders and update delivery status. Those functions were disabled before the refund limit was set and the passport issued.

ServiceNow describes the passport as a cryptographically signed authorization credential. Permissions are meant to be checked while an agent works, with controls tied to company policies, approvals and IT environment information. The company says documentation will support later reviews. Restrictions are intended to follow an agent across systems and through tasks handed to other agents.

Nvidia is a design and launch partner. ServiceNow says the passport uses open specifications and will connect credentials to infrastructure controls through Nvidia Open Agent Safety Platform. The demonstration showed the intended controls; how they work across customer environments remains to be seen. AI Passport is expected in the fourth quarter of 2026.

ServiceNow also plans EmployeeWorks features for building assistants without programming, and ToolSmith for developers to describe tools and integrations in ordinary language. Features to detect unapproved agents and consolidate AI costs are also expected in the fourth quarter. The company says these features have controlled availability and are not yet generally available.