ServiceNow gives AI agents a 'passport': signed rights, business context, and a kill switch
ServiceNow announced on 15 October 2026 AI Passport — a cryptographically signed attestation document intended to determine what each individual AI agent is actually allowed to do, verified at every step as the agent moves between workflows and tools. NVIDIA is the first design and launch partner, and the product is expected to be available in the fourth quarter of 2026 — a company expectation, not a shipped product. The launch lands in the middle of an open power question: who should govern agent authorization — an orchestration vendor like ServiceNow, or identity providers like Okta and Microsoft?
What was announced
AI Passport is described by ServiceNow as a new "authority layer" that gives businesses runtime control over what AI agents and MCP servers (Model Context Protocol) are authorized to do at any point in a multi-agent workflow. These are the company's own claims in the press release via Business Wire/Yahoo Finance.
According to Diginomica, which attended an analyst briefing with a demonstration, AI Passport is a cryptographically signed attestation document attached to each individual agent. It is intended to validate what the agent can and cannot do, and is verified at every step in the process.
The point is to replace trust in a third-party agent's own claims about its rights with a verifiable permissions record. According to the press release, each agent gets its own passport, issued by the Super Intelligence Control Tower and grounded in the company's live business context through the ServiceNow Configuration Management Database (CMDB), Governance, Risk and Compliance (GRC) policies, and workflow approvals.
Jon Sigler, executive vice president and general manager for the AI Platform at ServiceNow, frames the motivation this way: "As AI agents proliferate across the enterprise, visibility alone is not enough. Businesses need assurance that every AI agent operates within its authorized scope," he says in the press release.
Demonstrated in practice: the $25 refund limit
The most concrete demonstration from the briefing was simple: an administrator set a $25 limit for a refund agent. A refund request of $18 was automatically approved. A request of $40 was blocked from automatic execution (Diginomica).
The example shows the mechanism ServiceNow is betting on: business rules — here a financial threshold — are written into the agent's signed passport and enforced in real time, not afterward in an audit report.
A "dimmer", not just a kill switch
Despite the title of the Diginomica article, ServiceNow stresses that the governance is graduated. Nenshad Bardoliwalla, senior vice president for AI products at the company, says: "Real governance is graduated. It's more like a dimmer switch. You monitor, you pause, you restrict, and eventually, if necessary, you deactivate" (Diginomica).
The press release also refers to a machine-speed kill switch and runtime control at every point, but these are company claims in launch material, not independently verified.
The enforcement: NVIDIA and the infrastructure layer
The enforcement does not happen in ServiceNow itself, but in connected infrastructure. NVIDIA is the first design and launch partner and, according to the joint launch material, is building a production-ready implementation through the NVIDIA Open Agent Safety Platform, which includes OpenShell (an open-source runtime) and the NVIDIA Sentry security framework (Help Net Security).
Itay Ozery, director of product marketing for Networking and Cybersecurity at NVIDIA, justifies the choice of layer: "As enterprises move from experimenting with agents to running them in production, securing them requires enforcement at the infrastructure level," he says in Help Net Security.
The roles are divided as follows: ServiceNow's SI Control Tower acts as the policy hub and oversight layer, while the NVIDIA platform enforces the rules where the work is actually performed.
The positioning question: orchestration or identity?
Pat Casey, ServiceNow co-founder and CTO, distinguishes between two types of agents: those working under your own identity — in which case your permissions are already defined — and those running as administrator or under another identity. "If it's running as admin, or under a different identity than your own, it's much more likely to require central governance," he says in an interview with AIJourn. In that situation, "the agent can operate across organizational and security boundaries that would otherwise constrain what you can do."
But ServiceNow does not stand alone in this territory. According to AIJourn, Microsoft's Agent 365 and Okta's runtime gateway compete for a similar authorization domain. ServiceNow's differentiation argument rests on connecting an agent's access rights with the business rules governing the work — precisely the CMDB, GRC, and approval foundation ServiceNow already holds. This, however, is largely ServiceNow's own framing; it is not tested against the actual capabilities of Okta or Microsoft in the available material.
What remains unresolved
Several things remain open, and precision is warranted:
- Availability is a company expectation, not a product. ServiceNow expects AI Passport in the fourth quarter of 2026, alongside shadow AI detection and AI Cost Management. It has not shipped yet (AIJourn).
- "Open specifications" are not named. The launch material says Passport is built on open specifications intended to allow other governance products, identity providers, and enforcement points to issue and verify attestation documents (Diginomica) — but no concrete specifications are identified in the launch.
- The Veza question is unclear. Bardoliwalla confirmed at the briefing that Passport is not based on Veza (Diginomica), while AIJourn points to ServiceNow's acquisition of Veza, which maps existing permissions, as part of the platform's information foundation. The two statements can be reconciled — the acquisition exists, but Passport is not built on it — but the exact relationship is underdocumented.
- No pricing. None of the sources provide a price.
- The NVIDIA implementation has not been independently verified as production-ready, although this is claimed in the joint launch material.
For context: AI Passport was announced the same day ServiceNow expanded EmployeeWorks, its no-code platform for creating AI agents, and launched ToolSmith, a development environment the company says cuts agent tool-building from days to minutes (MSN).
Why it is worth watching
Technically, the core idea is simple: signed attestation documents tied to business context, verified at every step, with graduated governance. But the strategic stakes are larger. If ServiceNow succeeds in positioning SI Control Tower as the place where agent rights are defined, the company moves the boundary between identity and business logic — an area identity providers have traditionally held. The question of whether an orchestration vendor should actually be the authority deciding what an agent may do remains open, and the answer will likely be determined by which platform customers already trust when the rules are enforced.