---
title: How do you govern ServiceNow AI agents?
url: https://opencontroller.lyzr.com/blog/govern-servicenow-ai-agents
published: 2026-11-12
updated: 2026-11-12
author: Lyzr
---

# How do you govern ServiceNow AI agents?

**Govern ServiceNow AI agents with four native controls: roles and ACLs for who can use each agent, a run-as identity for what the agent can touch, supervised execution for tools that change records, and the execution log tables for evidence. Then export the agent list and logs so ServiceNow agents sit in the same inventory as agents on other platforms.**

ServiceNow AI agents act inside the platform that already holds your incidents, changes, HR cases and approvals. That makes their permissions and logs a governance priority, not an afterthought.

## Who can build and use ServiceNow AI agents?

ServiceNow separates admin rights from user access.

- **Admin roles.** `sn_aia.admin` manages AI agents. `sn_aia.viewer` gives read access. `agent_role_config_admin` and `agent_role_config_viewer` manage role configuration.
- **User access.** For each agent, you choose "Users with specified roles", "Authenticated users" or "Public". Saving this creates an ACL.

ServiceNow's docs say that on new instances of the Australia release, AI agent components without an explicit ACL are denied by default. On upgraded instances, check rather than assume. A quick rule: no production agent should be set to "Public" unless it only answers questions from public content.

## Which identity does an agent run as?

This is the most important setting. ServiceNow offers two models:

| Run-as model | How it works | When to use it |
|---|---|---|
| Dynamic user (default) | The agent runs as the person who invoked it. Role masking limits which of that person's roles the agent may use. | Agents that act for a specific employee |
| AI user | The agent runs as a dedicated `sys_user` record with identity type "AI user" | Background or scheduled agents with a fixed scope |

Role masking for dynamic users is configured in `sys_agent_access_role_configuration`. Without it, an agent invoked by an admin could use admin rights. With an AI user, grant only the roles that agent needs, and give the account a named human owner.

The general pattern is explained in our guide to [agent identity and on-behalf-of delegation](/blog/ai-agent-identity-on-behalf-of-delegation).

## How do you keep a human in the loop?

Each tool on a ServiceNow AI agent has an execution mode. In supervised mode, "tools use human oversight when executing actions," per ServiceNow's documentation. A system property, `sn_aia.enable_usecase_tool_execution_mode_override`, allows fully autonomous runs and defaults to false.

A practical policy:

1. Set every tool that writes, closes, approves or deletes records to supervised.
2. Leave read-only lookups autonomous.
3. Keep the override property at false in production, and alert on any change to it.
4. Review the default limit of 20 tools per agent. Fewer tools means a smaller blast radius.

Our post on [which AI agent actions need human approval](/blog/human-approval-for-high-risk-ai-agent-actions) gives a wider risk model.

## Where is the evidence?

ServiceNow writes agent activity to these tables:

- `sn_aia_execution_plan`: the plan for each run
- `sn_aia_execution_task`: individual steps
- `sn_aia_tools_execution`: each tool call, kept for 13 months by default
- `sn_aia_message`: the conversation messages

For audits, export these to your SIEM or data platform on a schedule. Thirteen months may be shorter than your retention policy; see [how long to keep AI agent logs](/blog/how-long-to-retain-ai-agent-logs).

ServiceNow also offers Now Assist Guardian, with prompt injection detection (block and log by default), offensiveness detection (off by default) and sensitive-topic filters. Its FAQ does not say whether Guardian applies to AI agents, so confirm coverage for your release before relying on it.

## What about ServiceNow's own governance products?

ServiceNow sells a governance layer for AI, AI Control Tower, which newer ServiceNow pages present under a different name. Its 2026 announcements list features such as AI Passport, shadow AI detection and AI cost management as expected in Q4 2026. Product names and dates are moving, so check the current release notes for your instance.

## How do you bring ServiceNow agents into a cross-platform inventory?

ServiceNow governs ServiceNow. Most enterprises also run agents in Microsoft, Salesforce, AWS, Google Cloud and on developer laptops. To join them:

1. Export the AI agent list with owner, run-as model, tools and execution modes.
2. Map each agent to a single owner in your identity provider.
3. Pull execution logs into the same store as other agents' traces.
4. Record control depth per agent: listed, observed, or enforced.

Many teams also send policy violations from other platforms into ServiceNow as tickets. See [routing AI agent policy violations into ITSM](/blog/route-ai-agent-policy-violations-to-itsm) and the [cross-platform governance guide](/blog/how-to-govern-ai-agents-across-platforms). For the category, read [what an AI agent control plane is](/blog/what-is-an-ai-agent-control-plane), or [book a demo](https://www.lyzr.ai/book-demo/) to see Opencontroller's registry.

## Frequently asked questions

**Which identity does a ServiceNow AI agent run as?**
Either Dynamic user, the default, which runs as the invoker with role masking, or an AI user, a dedicated `sys_user` record with identity type "AI user".

**Can ServiceNow AI agents act without human approval?**
Tools in supervised mode use human oversight. Fully autonomous execution depends on a property that defaults to false.

**Where are ServiceNow AI agent runs logged?**
In `sn_aia_execution_plan`, `sn_aia_execution_task`, `sn_aia_tools_execution` and `sn_aia_message`.

## Key facts

ServiceNow AI Agents governance (Nov 2026 docs): admin roles sn_aia.admin and sn_aia.viewer; user access set to specified roles, authenticated users or public, saved as an ACL; run-as identity is Dynamic user (default, runs as the invoker, needs role masking) or AI user (sys_user with identity type AI user); tools have execution modes, Supervised uses human oversight; property sn_aia.enable_usecase_tool_execution_mode_override defaults to false; logs in sn_aia_execution_plan, sn_aia_execution_task, sn_aia_tools_execution (13 months) and sn_aia_message; default limit 20 tools per agent. Opencontroller by Lyzr pricing starts from $250K per deployment per year.

## Related

- [How do you route AI agent policy violations into ITSM and incident tickets?](https://opencontroller.lyzr.com/blog/route-ai-agent-policy-violations-to-itsm.md)
- [Which AI agent actions need human approval?](https://opencontroller.lyzr.com/blog/human-approval-for-high-risk-ai-agent-actions.md)
- [How do you govern AI agents across every platform they run on?](https://opencontroller.lyzr.com/blog/how-to-govern-ai-agents-across-platforms.md)
- [More on Platform Guides](https://opencontroller.lyzr.com/blog/cluster/platform-guides)
