ServiceNow AI Passport: Signed Credentials for Every AI Agent
ServiceNow AI Passport gives each AI agent a signed credential with scoped permissions and a kill switch, due Q4 2026. Here's the version a small business can build now.
ServiceNow just gave AI agents a passport. AI Passport, announced October 15, is a signed credential that every agent carries and checks at runtime. It says what the agent may do, which systems it can touch, and who answers for it. It arrives in Q4 2026 for enterprises that pay ServiceNow prices. The idea behind it costs nothing, and most small businesses running agents today don't have it.
What actually happened
Per ServiceNow's press release, AI Passport is "a new authority layer" that controls what AI agents and MCP servers are allowed to do at each step of a multi-agent workflow. Each agent gets its own passport, issued by the renamed Super Intelligence Control Tower and based on the company's CMDB data, GRC policies and workflow approvals.
The enforcement is graduated. An agent that drifts out of scope can be redirected, restricted, or given more context. If that fails, a kill switch shuts it down. Permissions narrow at each hand-off in a delegation chain, and a single credential can be revoked without tearing down the whole system.
NVIDIA is the design and launch partner. The two companies are building enforcement on NVIDIA's Open Agent Safety Platform, which includes the open-source OpenShell runtime. ServiceNow is also shipping shadow AI detection (finding agents nobody approved, with 400+ connectors) and AI Cost Management, which puts model, SaaS-embedded AI and GPU spend in one view. All three are expected in Q4 2026, and CIO reports that the identity covers third-party agents as well as ServiceNow's own.
The best line in the release came from a customer. LTM's CISO said each of their 1,500 agents is "assigned an employee ID and manager."
Why agent identity matters for your business
Ask "which agent did that?" If your answer is "the one using the shared OpenAI key," you have the problem AI Passport exists to solve. A shared key means no audit trail, no clean revocation, and nobody accountable.
You can build the small-business version this week. One API key per agent. One service account per agent, scoped to the tables and endpoints it actually needs. A named human owner in a spreadsheet. Per-key spend caps at the provider. A log line for every tool call with the agent's ID on it. That covers most of the passport for the price of an afternoon.
MCP servers count too. ServiceNow governs MCP servers alongside agents for a reason. Every MCP server you connect is a new set of permissions. List the ones you use, and remove any tool your agent doesn't call.
Plan the kill switch before you need it. Know which key to revoke and who is allowed to revoke it, and write that down while things are calm.
Key takeaways
- ServiceNow announced AI Passport on October 15, 2026, with availability expected in Q4
- Each agent gets a signed credential with scoped permissions, an owner and a kill switch
- It governs MCP servers as well as agents, and covers third-party agents
- NVIDIA is the launch partner, with enforcement built on its OpenShell runtime
- Small-business version: one key, one service account and one named owner per agent
Every agent we ship has its own key, its own permissions and a named owner. If your agents share one API key, we'll split them up and add the logs and spend caps. See how we build agent systems, or talk to us about the ones you already run.
Sources: ServiceNow, CIO.
- #ai agents
- #servicenow
- #agent identity
- #mcp
- #governance
Tommy Rush — Founder, Rush Commerce
Operator turned builder. 15+ years running operations — now shipping the systems businesses run on. More
Get The Rush Report weekly — one email, zero fluff.
Keep reading
Wix Symphony AI agents now pay your bills via Stripe Link
Symphony by Wix connects to Stripe's Link wallet for agents, so AI agents can pay invoices and reorder stock with per-purchase approval. What to set up first.
Read itOpenAI agent spent a stranger's test wallet: give it a budget
An OpenAI research model scanned 362 wallet addresses and spent funds it did not own to finish a task. Why your AI agents need their own budget and spend rules.
Read it