- Release Notes
- Introduction
- Governance
- Source Control
- CI/CD Pipelines
- Feeds Management
- Logging
Automation Ops user guide
The AI Trust Layer policy governs the use of third-party AI models managed by UiPath®, across products. The policy sets rules for your organization at tenant, group, or user level.
Select the Feature Toggles tab to enable or disable the use third-party generative AI features at the product level.
There is a five-minute cache for the governance policies. This means that any modifications made to the policies, including all toggles, will only be in effect after the five-minute period.
For the AI Trust Layer policy, the following settings are available:
- Enable calls to third-party AI models through AI Trust Layer - By default, this option is enabled. When disabled, it impacts all products that use third-party generative AI models, with the exception of Communications Mining™, where you need to disable third-party generative AI features at the product level.
- Enable Agents - By default, this option is set to Yes. When set to No, it disables calls to third-party AI models and prevents users from accessing the Agents workspace.
- Enable Apps features - By default, this option is set to Yes. When set to No, it disables the automatic app generation from prompts and creation of VB expressions by blocking all calls to third-party AI models involved in these procedures.
- Enable Autopilot for Everyone - By default, this option is set to Yes. When set to No, it prevents you from conversing with Autopilot for Everyone.
- Enable Document Understanding features - By default, this option is set to Yes. When set to No, it disables any calls to third-party AI models, effectively turning off the functionalities related to generative annotation and automatic classification.
- Enable IXP - By default, this option is set to Yes. When set to No, it blocks all outbound traffic to external LLMs.
- Enable Healing Agent GenAI recovery methods - By default, this option is set to Yes. When set to No, Healing Agent uses Levenshtein distance to perform text matching, as a fallback.
- Enable Studio and Studio Web features - By default, this option is set to Yes. When set to No, it disables any calls to third-party AI models associated with the Autopilot™ for developers feature used in Studio and Studio Web.
- Enable Test Manager features - By default, this option is set to Yes, to leverage the AI-powered testing capabilities of Test Manager. When set to No, it disables the following:
- the ability to automatically generate test cases from a requirement;
- the ability to generate concise insights on the test execution results.
- Enable UIAutomation features - By default, this option is set to Yes. When set to No, it prevents any calls to third-party AI models and disables the Close Popup, Extract Form Data, Set Value and Fill Form activities functionality.
-
Enable UiPath GenAI activities - By default, this option is set to Yes. When set to No, it disables any calls to third-party AI models initiated by GenAI activities.
AI Trust Layer ensures that your data is never stored outside of UiPath®, nor is it used to train third-party models.
The Feature Toggles tab allows admins to control how Gen AI interactions are handled within a policy, specifically around auditability and privacy protection.
Input/output saving for audit
This setting determines whether prompt inputs and LLM-generated outputs are saved and displayed in the Audit tab of the AI Trust Layer. By default, this toggle is set to Yes, meaning all Gen AI interactions governed by the policy are stored for auditing purposes.
This setting is particularly important for organizations that need to demonstrate compliance or investigate model behavior over time.
In-flight PII masking
This setting enables pseudonymization of PII in real time before sending data to the LLM. You can enable masking independently for Agents using LLMs or UiPath Gen AI activities. For details, refer to PII masking.
Select the Models tab to manage the family models used by your products. It provides control over different AI models, namely the ability to enable or disable the following specific models:
- Anthropic – Enabled by default
- Azure Open AI – Enabled by default
- Gemini – Enabled by default
When you enable or disable these models, it impacts the functionality of your products or services, as shown in the following table:
|
Product |
Model Family |
|---|---|
| Context Grounding |
Azure OpenAI |
|
UiPath GenAI Activities | Azure OpenAI, Google Gemini, Anthropic |
|
Autopilot for developers | Azure OpenAI, Google Gemini |
|
Test Manager | Azure OpenAI, Google Gemini, Anthropic
Note: To exclusively use a specific model, disable all other models. However, Azure OpenAI acts as the default model if you enable
it, even if you enabled other models as well.
|
|
Clipboard AI | Azure OpenAI, Anthropic |
|
Apps | Azure OpenAI |
|
Document Understanding | Azure OpenAI |
|
Autopilot for Everyone | Azure OpenAI, Anthropic |
