ServiceNow launches a signed credential designed to govern what AI agents may do
The company is launching AI Passport on October 15, 2026: a signed credential that agents must present in real time. But enforcement, by the company's own account, is still being built – and the passport is only as trustworthy as the data it is issued from.
On October 15, 2026, ServiceNow launched AI Passport, a cryptographically signed credential that AI agents are to reference at runtime to verify what they are authorized to do. The passport covers both first-party and third-party agents as well as MCP servers across multi-agent workflows, and replaces, according to the company, self-declared permissions with verifiable identity and capability attestation (Help Net Security; Forkast via Yahoo Tech).
The launch comes alongside a broader expansion of what the company calls the Super Intelligence Control Tower, with shadow AI detection and AI Cost Management. All of it is, according to the company, expected to be available in the fourth quarter of 2026. The question the launch raises is whether a signed credential can actually be enforced faster than the company's own timelines suggest – InfoWorld reports that the enforcement implementation is still being built.
How it works
An AI passport is issued by the Super Intelligence Control Tower and anchored in ServiceNow's CMDB (configuration management database), GRC policies, and workflow approvals. When an agent is to perform an action, it must present cryptographically signed proof of identity and a capability attestation before it is allowed to act – rather than simply asserting what it is permitted to do (Forkast/Yahoo Tech).
This also applies to third-party agents and MCP servers – an increasingly common building block in agent architectures. That gives enterprises a control point around agents they did not build themselves, across entire multi-agent chains.
The specifications are, according to the company, open, so that other governance products, identity providers, or enforcement points can issue and verify the same type of credential (InfoWorld).
The demo: $18 yes, $40 no
ServiceNow demonstrated the mechanism with a third-party customer service agent configured for refunds under $25. An $18 refund went through automatically. A $40 request was rejected at runtime because it exceeded the threshold for human review (SiliconANGLE).
Nenshad Bardoliwalla, SVP of AI products at ServiceNow, described the enforcement as a sliding scale rather than a binary switch: "It's more like a dimmer switch," he said, according to SiliconANGLE. "You monitor, you pause, you restrict, and eventually, if necessary, you deactivate."
In other words: from passive observation, via temporary stops and narrowed permissions, all the way to an actual "kill switch."
NVIDIA as partner
NVIDIA is named as design and launch partner and is building a production-ready enforcement implementation on its Open Agent Safety Platform – which includes OpenShell, an open-source runtime environment for secure execution, and the Sentry security framework (Help Net Security). Among other things, the passport is to be verifiable in such environments at runtime.
The rest of the package
Announced at the same time, all with the company's Q4 2026 as the target:
- Shadow AI detection – mapping agents the enterprise has not itself registered. NTT Data says it is already using AI Control Tower against shadow IT, and that the company is beginning to adopt EmployeeWorks, Jamie Robins, head of the company's ServiceNow practice, told SiliconANGLE.
- AI Cost Management – cost visibility into AI usage.
- AI Agent Studio – lets employees build agents with governance built in by default (Forkast/Yahoo Tech).
- ToolSmith – a natural-language development environment for automation in ServiceNow EmployeeWorks. ServiceNow states that it is trained on more than 75,000 APIs and optimized for reliability and token efficiency (InfoWorld).
The caveats
Two conditions are flagged as decisive by analysts, and both warrant a skeptical reading of the demonstration.
The enforcement is unfinished. InfoWorld reports that ServiceNow says the enforcement implementation is still being built. That means the $18/$40 demonstration cannot be read as a guarantee of how general availability will behave.
The data behind the passport. Analyst Frank Dickson notes in a comment to InfoWorld that a credential is only as trustworthy as its issuer: "A passport is only as trustworthy as the office that issues it. The issuing office here is ServiceNow's CMDB, its GRC policies, and its workflow approvals. If that data is stale, the passport is stale." He adds that open specifications only matter if others adopt them – NVIDIA is so far the launch partner. Forkast additionally points out that effectiveness depends on the granularity of the credential issuance itself: a coarse-grained passport yields coarse-grained control (Forkast/Yahoo Tech).
What remains
The message is concrete: a signed credential, checked at runtime, with a graduated enforcement scale and a major partner on board. But what we know so far rests on secondary reporting of the company's own claims – the availability date, the partnership terms, and the ToolSmith figures are all company statements. What remains to be seen is whether the open specifications gain real adoption beyond NVIDIA, whether the enforcement lands by Q4 2026 as promised, and whether enterprises' CMDB and GRC data are clean enough for the passports to be issued on a sound basis. The answers will determine whether this becomes a genuine governance primitive – or an announcement ahead of its implementation.