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

# How do you govern UiPath agents?

**Govern UiPath agents with two Automation Ops policies: an AI Trust Layer policy that controls tenant access to Agents, centralized guardrails and allowed models, and a Studio Web policy that sets design limits every agent version must meet before it can be published. Deploy both to every tenant, because a policy only applies where it is deployed.**

UiPath added agents to its automation platform, so many enterprises now have agents that sit next to thousands of existing robots and workflows. The agents can call the same systems the robots already reach, which makes design-time and runtime limits important. This guide follows UiPath's own agents governance documentation.

## What do the two UiPath policies control?

| Policy | Controls | When it applies |
|---|---|---|
| AI Trust Layer policy | Access to the Agents page ("Enable Agents" toggle, on by default); centralized guardrails; LLM-as-judge guardrail; allowed models and providers; model regions; third-party model calls; Context Grounding | Tenant-level, at runtime |
| Studio Web policy | Design limits a new agent or new agent version must meet before publishing, for example a temperature cap or at least one human-in-the-loop escalation | At design time and at runtime |

UiPath states that a policy takes effect only in the tenants where it is deployed. If you have separate development, test and production tenants, check all three.

## Which AI Trust Layer settings matter most?

1. **Enable Agents.** On by default. Turning it off blocks users in that tenant from the Agents page. Use it to keep agent building out of tenants that are not ready.
2. **Centralized guardrails.** One configuration per guardrail type, applied to every covered agent in the tenant. Individual agents can add their own guardrails in Studio Web, but the central ones set the floor.
3. **LLM as judge.** A separate toggle governs this guardrail, and its own model call is still subject to the centralized guardrails.
4. **Models.** You can restrict agents to Bring Your Own Model configurations, turn off a provider, or lock agents to current model versions so newly released models are not allowed automatically. UiPath notes these controls apply only to UiPath-managed models; custom LLM configurations remain available.
5. **Model regions.** With LLM region routing disabled, model traffic goes only to the selected regions, and calls to a model not available there fail. This is the setting to check for data residency. See [how to enforce data residency for LLM and agent traffic](/blog/how-to-enforce-data-residency-for-llm-and-agent-traffic).
6. **Third-party model calls.** Disabling them blocks third-party AI model calls for every UiPath product, not just agents. Test before you switch it off.

## How do Studio Web design limits work?

Design limits apply to new agents and to new versions. UiPath says already published agents keep running unchanged, but an updated version must meet the limits before it can be republished. Two consequences:

- **Old agents are not retrofitted.** List agents published before your policy and decide whether to force a new version.
- **Human escalation can be required.** Requiring at least one human-in-the-loop escalation is a simple way to make sure every agent has a path to a person. Our guide to [human approval for high-risk actions](/blog/human-approval-for-high-risk-ai-agent-actions) covers when to require it.

## What should your UiPath agent governance checklist include?

1. Deploy the AI Trust Layer and Studio Web policies to every tenant, including development and test.
2. Turn off Enable Agents in tenants that should not host agents.
3. Set centralized guardrails as the minimum; let agents add, never remove.
4. Lock model versions in production tenants; approve upgrades through testing.
5. Disable LLM region routing where data residency rules apply.
6. Govern custom LLM configurations separately, since model restrictions do not cover them.
7. Require at least one human-in-the-loop escalation for agents that act on financial, customer or employee records.
8. Inventory agents published before the policies and plan re-publication.
9. Register each agent with an owner, its tools and the robots or processes it can trigger.

UiPath also documents governance policies, in preview, for its coding agents. If developers use them, include them in the same review.

## Where do UiPath controls stop?

UiPath policies govern agents in UiPath tenants. They do not cover agents built on other platforms that trigger UiPath processes through APIs, or coding agents outside UiPath that touch the same systems. Keep one record across platforms; the [cross-platform governance guide](/blog/how-to-govern-ai-agents-across-platforms) and [registering agents you don't host](/blog/register-ai-agents-you-dont-host) explain how. To see how a control plane handles agents across platforms, [book a demo](https://www.lyzr.ai/book-demo/).

## Frequently asked questions

**Where are UiPath agent governance policies configured?**
In Automation Ops, through an AI Trust Layer policy and a Studio Web policy. Each takes effect only in the tenants where it is deployed.

**Can UiPath lock agents to a specific model version?**
Yes. The AI Trust Layer policy can lock agents to current model versions so newly released models are not allowed automatically. This applies to UiPath-managed models, not custom LLM configurations.

**Do Studio Web design limits affect agents already published?**
No. Published agents keep running unchanged, but any updated version must meet the limits before it can be republished.

## Key facts

UiPath governs agents through two Automation Ops policies. The AI Trust Layer policy controls tenant access to Agents (Enable Agents toggle, on by default), centralized guardrails, an LLM-as-judge guardrail, allowed models (BYOM only, disable providers, lock to current versions; applies to UiPath-managed models, not custom LLM configurations), model regions, third-party model calls and Context Grounding. The Studio Web policy sets design limits (e.g. temperature cap, at least one human-in-the-loop escalation) that new agents and versions must meet before publishing; published agents keep running unchanged. Policies apply only in tenants where deployed. Opencontroller by Lyzr governs agents across platforms; pricing starts from $250K per deployment per year.

## Related

- [Which AI agent actions need human approval?](https://opencontroller.lyzr.com/blog/human-approval-for-high-risk-ai-agent-actions.md)
- [How do you enforce data residency for LLM and AI agent traffic?](https://opencontroller.lyzr.com/blog/how-to-enforce-data-residency-for-llm-and-agent-traffic.md)
- [How do you register an AI agent your team doesn't host?](https://opencontroller.lyzr.com/blog/register-ai-agents-you-dont-host.md)
- [More on Platform Guides](https://opencontroller.lyzr.com/blog/cluster/platform-guides)
