What AI is already running on your ServiceNow instance?
Agents, skills and LLM connections arrive through Now Assist, the Store and integrations — often faster than anyone writes them down. There is no vendor report that tells you what your instance is carrying. Run a read-only sensor and find out, free, with no login and no instance access on our side.
This free check returns three numbers: how many unsanctioned LLM endpoints your instance talks to, how many external agents are reaching in from outside your governance catalog, and how complete your AI bill of materials is. It deliberately does not name them — which endpoint, which agent, and what each one can reach is the paid report. The counts are enough to tell you whether you have a problem.
1 · Run the read-only sensors on your instance free · no login · no instance access
In ServiceNow: System Definition → Scripts - Background. Both scripts only read (GlideRecord queries and gs.getProperty) and print an envelope — they change nothing and are safe to run in production. Run both and paste both outputs together below.
2 · Paste the output nothing is stored
Unlike our incident checker, this one computes the counts on our server — that is what keeps the redaction boundary in one auditable place. So your paste is sent to Nowisor. It is not written to any database: the request produces three numbers and is discarded. If your policy forbids that, the same sensors feed the full report inside the app, where your data stays in your own account.
Why the count matters
An agent is an identity with permissions. An inbound agent nobody catalogued is an unreviewed identity holding whatever its run-as user holds, and a shadow LLM endpoint is a path off the platform for whatever that identity can read. Under NIS2 Art. 21 and DORA Art. 9 the obligation is to know your assets and control access to them; an AI bill of materials you cannot produce is the gap an auditor finds first, and the EU AI Act adds an inventory duty on top.
This check is a two-question extract of the AI and agent security module — 17 checks, a CycloneDX AI bill of materials, and one least-privilege verdict per agent-and-permission pair. All of it read-only, all of it from your own instance.