USE CASES / SERVICENOW
Where a customer relies on ServiceNow, GovernorAI's job is not to duplicate it. It is to inventory the AI surfaces on the instance, to govern the consequential actions that happen outside the native workflow, and to keep an evidence layer that belongs to you and travels — one record for every governed agent action, whichever platform it ran on. This page is explicit about which of those happens inline and which does not.
Three ways GovernorAI touches ServiceNow, and they carry different powers. Only gateway_execute mediates the request body; on the native surface a deny means a downstream control action, not an inline refusal.
DISCOVERY · READ-ONLY
The connector authenticates with an OAuth2 client-credentials grant — long-lived Basic credentials are still accepted for legacy connections — and reads a small set of tables through the Table API. Nothing is written and nothing is intercepted. The output is an inventory of the AI surfaces that exist, what they touch, and a coarse risk read on each.
sys_cb_topic Virtual Agent topics Active conversational topics — the surfaces an employee or customer actually talks to.
active = truesys_nlu_model NLU models The natural-language models behind those topics, listed with the intents they resolve.
active = truesys_pd_model → ml_solution Predictive Intelligence Predictive models, read from the legacy table where it is present and from the newer solution table where it is not. The fallback is explicit, not silent.
two-table readsys_hub_flow Flow Designer flows Active flows filtered to AI-related names and scopes — Now Assist, generative, predictive, AI Search and similar.
plugin-dependentTwo of these reads can come back empty for reasons that are not "nothing is there". Flow Designer tables require the Flow Designer plugin to be active; when the instance answers that the table type is invalid, the connector logs it and skips that surface rather than reporting zero flows as a finding. And the flow sweep is keyword-filtered, so an AI-driven flow named in a house convention we do not recognise will not be picked up. Discovery narrows the unknown; it does not certify the instance.
THE BOUNDARY, STATED PLAINLY
This is the part of a ServiceNow conversation that goes wrong when a vendor blurs it. GovernorAI ships an enforcement point capability registry that declares, per enforcement point, which interaction kinds it can inspect and which outcomes it can truthfully execute — and it fails closed if a richer outcome is asked of an enforcement point that cannot carry it. Here is what that registry says about each of the three.
seam: saas_native scoped app → proxy check seam: gateway_execute allow · deny — and nothing richer. A shaping outcome asked of this enforcement point fails closed.allow · deny · pause — a paused step carries the reason and the policy that held it.allow · deny · pause · redact · mask · constrainThe in-SaaS enforcement point is genuinely useful — it is how the inventory stays current and how an incident gets a timeline — but it is an out-of-band surface, and treating it as an inline control would be a false claim about where GovernorAI sits. The inline control is real in the other two lanes. Both statements ship in the same registry, and the registry is exposed to operators as a read-only API rather than described only in marketing.
The in-flow check is a real, tested Go backend: a per-agent shared secret, constant-time authentication, fail-closed when the evaluator is not wired, and an audit event emitted whether the answer was allow, deny or pause. The ServiceNow-side half — the scoped app, its connection record and the flow action that calls out — is source that cannot be validated by our own test suite; it needs a live ServiceNow instance to exercise end to end. Treat that half as an integration to prove in your environment, not as a shipped result.
CONNECTION MODEL
Most connectors in GovernorAI do one thing. ServiceNow is one of only two platforms — Salesforce is the other — where a single connection can carry both read-only discovery and governed execution, behind one record with independent capability flags. Everything else in the catalogue is discovery-only and does not need the facade.
Periodic agent discovery against the Table API. Read-only, and the low-friction way in: an inventory arrives before anyone is asked to let a policy decide anything.
Enabling governance for a discovered topic registers it and issues a per-agent secret. Until a real governed-execution record exists and is linked, the connection is not allowed to report itself as governing.
The invariant is enforced, not documented. A connection cannot sit in an active, governing state unless the underlying governed-execution record actually exists and is linked to it — and a discovery connection that is unhealthy cannot be promoted into governance until it is repaired. Turning a capability off goes through the same lifecycle as turning it on, so the record never claims a posture the subsystem is not in.
The two capabilities are separate flags on one connection. That is the enterprise adoption path the rest of the product is built around: start read-only, establish an evidence trail, and add enforcement where the consequence justifies it — per topic, opted in one at a time, rather than as a platform-wide switch.
POSITIONING
ServiceNow is a broad platform with its own AI governance surface, and organisations that have standardised on it should keep using it. GovernorAI is deliberately outside the delivery stack — which is worth something in exactly three places.
An agent that reads a ticket and then moves money, writes to a clinical record or changes a production system is acting somewhere else. GovernorAI governs that call at its own action boundary, under the same policy set.
The same policy that governs a ServiceNow-adjacent action governs a Bedrock action group and an MCP tool call. The decision is made in one place, so the answer does not depend on which stack a team happened to use.
A hash-chained, tamper-evident ledger of governed actions, mapped to controls, held outside the systems being governed. If a platform choice changes, the evidence chain does not have to be re-argued.
GovernorAI's independence is the product. A governance record produced by the same platform that builds and runs the agent answers a different question than one produced by a party with no stake in the outcome — and the second is the one a security review, an auditor or a regulator is asking for. That is a statement about layer and posture, not about what any other platform can or cannot do.
Nothing on this page should be read as a comparison of framework coverage, or as a claim that any framework is satisfied on its own. Evidence is mapped to controls for the actions GovernorAI actually governed; actions on enforcement points you have not enabled produce no record, and an empty ledger for a surface means it was not governed rather than that nothing happened there.