Short answer: on October 15, 2026, ServiceNow announced AI Passport, which it describes as an authority layer giving enterprises runtime control over what AI agents and Model Context Protocol (MCP) servers may do. Each agent receives a cryptographically signed credential that is checked as it works. It is a meaningful market signal that agent governance is becoming a product category, but it is a vendor announcement with availability expected in the fourth quarter of 2026, not a settled standard and not something already deployed everywhere. The lasting lesson is vendor-neutral: agents need identity, bounded permissions, rules checked at the moment of action, and evidence of what happened.
What ServiceNow Announced
According to the ServiceNow press release of October 15, 2026, AI Passport gives each agent its own passport, issued by its Control Tower and grounded in configuration data, GRC policies and workflow approvals. It is built on open specifications so that third-party governance products, identity providers and enforcement points can issue and verify credentials. When an agent drifts from its scope, enforcement can redirect, restrict or reconnect it, or trigger a kill switch.
The announcement also lists shadow AI detection for unapproved agents, more than 400 connectors for AI discovery, access mapping based on actual entitlements, continuous observability of agent behaviour, AI cost management, and graduated alert, throttle and block responses. NVIDIA is named as a design and launch partner on an enforcement implementation. ServiceNow executive Jon Sigler is quoted as saying that “visibility alone isn’t enough.”
Watch: Who Is Governing Your AI Agents?
Is Agent Governance the Next GRC Category?
Traditional GRC works through policies, questionnaires, periodic assessments and retrospective audits. Those methods assume that humans take the actions and that there is time to review them. AI agents that read data, call tools and trigger transactions change the assumption. By the time a quarterly review finds a problem, an agent may have repeated it thousands of times.
That is why governance is shifting towards the moment of execution, the same direction we described in AI governance at machine speed and policy as code. A category is forming around agent identity, permissions, runtime enforcement and audit evidence. Whether it becomes a standalone market or folds into existing identity, security and GRC platforms is still open. Enterprises should expect several approaches and ask how any of them connect to their own policies and internal controls.
The idea of agents as identities is not new in our coverage. See AI agent identity and access governance and treating agents as non-human privileged users. What the announcement adds is a market signal that vendors are packaging it.
The Governance Chain: From Agent Identity to Audit Evidence
Whichever tooling you choose, the governance logic follows one chain. Each link needs an owner and a test.
Agent Identity
Every agent, whether built in-house or supplied by a vendor, has its own verifiable identity and a named human owner. You cannot govern what you cannot identify, so discovery of unapproved agents comes first.
Authorised Actions
Define what each identity may do: which systems, data and transactions it can reach, within what limits. Permissions are granted to the task, not to the model.
Control Rules
Translate policy into rules that can be checked at the moment of action, such as approval thresholds, segregation of duties and data-handling limits.
Exceptions
When an agent steps outside its scope, the event is flagged immediately and graded by risk. The response may be an alert, a throttle or a block, depending on the control.
Human Intervention
High-impact exceptions go to an accountable person who can approve, redirect, restrict or revoke. Define the escalation path and the time limit in advance.
Audit Evidence
Every identity, permission, action, exception and decision is recorded in a form an auditor can reconstruct later, linking the agent’s behaviour back to the policy and control it was meant to satisfy.
The chain is only as strong as its weakest link. Permissions with no audit evidence cannot be proven, and evidence with no human intervention path cannot change an outcome. The need for AI agent audit trails and the assurance gap in agent audit trails explain why the final link matters to internal audit.
An illustrative example
This is a hypothetical scenario, not a customer case study. A procurement team deploys an agent to draft purchase orders. Its passport allows it to read approved supplier records and create drafts up to a set value, and nothing else. One day it tries to change a supplier’s bank details. The control rule blocks the action, the exception goes to the procurement controller, who rejects it, and the event, the rule and the decision are logged against the original policy. Internal audit later samples these records to test that the rule works. Without the chain, the change might have gone through and been found at month-end, if at all.
Where to Start Without Waiting for a Product
Start with an inventory of the agents and AI services in use, as in AI inventory before AI governance. Assign an owner to each, map what each can access, and decide which actions need human approval. Include third-party agents, since vendors increasingly embed them, as discussed in AI agent third-party access as a board risk. Then test the controls, using the approach in auditing the agentic AI governance gap.
Crest.Digital positions Agentic GRC as a governance layer that connects AI actions to enterprise policies and internal controls, the same chain from identity to evidence, with human intervention built in. Our wider view is in agentic execution across the risk lifecycle, and third-party agents are covered by the TPRM platform.
Talk to Crest.Digital about governing AI agents through identity, control rules, exceptions and audit-ready evidence.
Questions for Your CIO, CISO and CRO
- Inventory: How many AI agents operate in our business, and who owns each?
- Permissions: What can each agent do, and who approved it?
- Exceptions: What happens, and how fast, when an agent steps outside its scope?
- Evidence: Could we reconstruct what an agent did last month for an auditor?
Frequently Asked Questions
According to ServiceNow’s October 15, 2026 announcement, AI Passport is an authority layer that gives enterprises runtime control over what AI agents and Model Context Protocol (MCP) servers may do. Each agent receives a cryptographically signed credential that is checked at runtime, and enforcement can redirect, restrict or revoke an agent that drifts from its scope. It is expected to be available in Q4 2026.
It is a credible candidate. As agents execute workflows rather than only draft text, governance has to operate during execution, not only through policies, questionnaires and retrospective audits. A single vendor announcement is a market signal, not proof of a settled category, and buyers will still compare approaches.
Model governance focuses on how a model is built, tested and monitored. Agent governance focuses on what an agent is allowed to do in your systems: its identity, permissions, actions and the evidence of what it did. Enterprises need both.
Build an inventory of the agents and AI services in use, assign owners, map the systems and data each can reach, and write down which actions need human approval. A tool can enforce those decisions, but it cannot make them for you.
Internal audit should test whether agent permissions match policy, whether exceptions are escalated and resolved, and whether the evidence trail is complete. Risk teams define the control rules and tolerances. Neither role is replaced by the tooling.
Source: ServiceNow Newsroom, October 15, 2026. This article is Crest.Digital’s independent commentary and is not affiliated with ServiceNow or NVIDIA.