ServiceNow pitches enforceable permissions for AI agents

ServiceNow has announced controls designed to limit what AI agents can do across enterprise systems, alongside new tools for employees to build assistants and automate routine work. The key idea is an agent credential that carries its identity, permissions and accountable owner, with checks applied as the agent acts.

Watch Desk analysis

What happened

SiliconANGLE reports that AI Passport, part of ServiceNow’s AI Control Tower, is designed to carry those rules across agents, tools and vendors. In a demonstration, a customer-service agent was allowed to issue refunds below $25: an $18 refund went through automatically, while a $40 request was blocked for human review.

The company also described graduated controls, from pausing an agent or restricting its tools to revoking its credentials. It says AI Control Tower is adding shadow-AI detection, more than 400 vendor connectors and spending oversight. AI Passport, shadow-AI detection and AI Cost Management are due later this quarter; the new EmployeeWorks assistants, routines, ToolSmith and Conversations API are in controlled availability. Operational Resilience for financial services is available now.

EmployeeWorks updates include job-specific assistants, background routines for repeatable tasks, and ToolSmith, which ServiceNow says can build tools and integrations from natural-language instructions. The company says ToolSmith is trained on more than 75,000 APIs. SiliconANGLE also reports that pricing combines seat licences with AI credits called assists.

Why it matters

As agents gain the ability to use business systems, a policy that merely says “please behave” is not much of a control. The refund example makes the proposed distinction clear: an agent can handle a permitted low-value task, while a request outside its authority is stopped for human review. If the credentials and checks work across vendors as ServiceNow intends, organisations could set limits at the point where an agent acts, rather than relying only on monitoring after the fact.

The same announcement pairs those controls with tools that make it easier for employees to create assistants and automate work. That is a useful tension to keep in view: easier agent-building increases the importance of clear permissions and oversight. The article describes a demonstration and product plans, not evidence of how these controls perform across real deployments.

Our read

The standout here is not another assistant with a cheery name. It is the attempt to make an agent’s permissions travel with it, and to reject an out-of-bounds action before it happens. That is a more practical governance proposition than hoping every organisation can keep an accurate list of agents and remember to check it.

For businesses considering the products, the questions are whether the rules hold across third-party agents and systems, and how administrators can audit decisions and exceptions. A guardrail that can be switched off, or that stops at the vendor boundary, is rather less reassuring.

What to watch

• When AI Passport, shadow-AI detection and AI Cost Management become available.
• Which third-party agents, identity providers and enforcement systems can issue or verify credentials.
• Whether customers can audit blocked actions, exceptions and changes to an agent’s permissions. Activity teaser: ServiceNow’s AI Passport is designed to give agents a signed identity and defined permissions, then check those limits as they act. In the company’s demonstration, an $18 refund was approved automatically while a $40 request was stopped for human review. The announcement also includes new tools for building employee assistants and automations, some of which are in controlled availability. Can portable, enforceable permissions make AI agents safer to deploy, or will organisations still struggle to keep control once agents span multiple vendors?

Discussion spark: Should enterprise AI agents be allowed to complete low-risk tasks automatically when their permissions are enforceable, or should a person approve every consequential action?

Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.