Automation Cloud Admin Guide

Last updated Dec 10, 2024

Managing AI Trust Layer

Managing AI Trust Layer policies

The Settings tab on the AI Trust Layer page allows you to manage third party AI models usage for your organizations, tenants, or user groups through AI Trust Layer policies. This helps you control user access to Generative AI features and ensures appropriate governance across your organization.

You get an overview of all active policies and their current statuses. In addition to this, you can view and manage policies and their deployments, as follows:

When you select the policy name, you are redirected to the respective AI Trust Layer policy in Automation Ops™ > Governance. Here you can view the policy details and, if necessary, make changes. For details, refer to Settings for AI Trust Layer Policies.
When you select Manage Deployments, you are redirected to Automation Ops™ > Governance > Deployment, where you can review all your policy deployments. For details, refer to Deploy Policies at Tenant Level.

Governing contextual data for GenAI features

The Context Grounding tab on the AI Trust Layer page allows you to manage and govern the data you use as context with UiPath® GenAI features. For more details about context grounding, see About Context Grounding.

You can check the Context Grounding indexes available across specific tenants. Data about the index data source, update time, and last query time is also available.

Moreover, you can perform the following operations:

Add a new index. For details, refer to Adding a new index.
View details for the index attributes, using the See more actions menu.
Permanently delete the index, including all embeddings and data representation within the index. You can delete the index form the See more actions menu.

Note:
If GenAI Activities or Autopilot for Everyone uses an index that is deleted, you must reconfigure them with a new index, otherwise they will fail.
Sync the index to update it with the most recent data from the data source. The sync operation overwrites the embeddings and captures only the data currently available in the data source. You can sync the index from the See more actions menu.

Adding a new index

To add a new index that you can use as data context with UiPath GenAI features, take the following steps:

On the AI Trust Layer page, go to the Context Grounding tab, then select the Add new button.
Provide the index name and, optionally, a short description.
Specify the source of the data that you plan to ingest and make queryable.
- If you choose Storage bucket as your data source, take the following steps:
  1. Specify the Orchestrator folder where the storage bucket resides.
  2. Indicate the storage bucket where the desired documents are available.
  3. Select the format of the files you want to index. If you want to ingest multiple file formats, you must select the All option.
    
    Note:
    You cannot change the selected option after creating the index.
- If you choose IS Connection as your data source, take the following steps:
  1. Specify the Orchestrator folder where the storage bucket resides.
  2. Select the connector and the connection. If no connection exists, you can add one.
  3. Specify the location of your documents.
  4. Select the format of the files you want to index. If you want to ingest multiple file formats, you must select the All option.
    
    Note:
    You cannot change the selected option after creating the index.
Create the index by selecting Sync now. Initially, the index status is Sync in Progress and changes when the sync is complete.

Viewing audit logs

The Audit tab on the AI Trust Layer page offers a comprehensive view of AI-related operations, with details about requests and actions, the products and features initiating requests, as well as the used models and their location. You can monitor all AI-related operations and ensure their compliance with your established guidelines and policies. Note that you can view log entries created in the last 60 days.

The audit data is displayed as a table, with each of its columns providing a specific information about the AI-related operations:

Date (UTC): This displays the exact date and time,when each operation was requested. It allows for accurate tracking of requests according to their chronological order, facilitating timely audits.
Product: The specific product that initiated each operation. This visibility allows tracing any operation back to its originating product for enhanced understanding and accountability.
Feature: The specific product feature that initiated the operation, facilitating issue traceability to particular features, if any occurred.
Tenant: The specific tenant within your organization that initiated the operation. This insight enables a more detailed overview and helps recognize patterns or issues at the tenant level.
User: The individual user within the tenant who initiated the operation. It allows for tracing activities at a granular user level, enhancing the oversight capabilities.
Model Used: The specific AI model employed to process each operation. This insight provides a better understanding of which models are handling which types of requests.
Model Location: The location where the used model is hosted. This information can assist potential troubleshooting or audit requirements that could arise from model performance in specific locations.
Status: The status of each operation—showing if it was successful, failed, or blocked. This quick way of identifying operational issues is crucial for maintaining a smooth, efficient environment.

Additionally, the filtering capability allows you to narrow down your audit based on criteria such as the date, product, used model, or status.

Furthermore, when you select an entry from the Audit table, you can access a Details section for a more in-depth review.

Handling PII and PHI data in audit logs

If you use the GenAI features, you must be aware that your audit logs might include Personally Identifiable Information (PII) and Protected Health Information (PHI). These details can appear in logs when processing documents or managing input prompts through both attended and unattended automation. You can view the input and output prompts in the Details section when you review specific requests.

The information that can contain PII and PHI includes user and product prompts sent to LLM models as well as the responses generated by these models.

You can track the origin of the PII or PHI in your logs by reviewing the request timestamps, input content, and associated metadata such as Action ID, Tenant, Product, Feature, and User.

If your compliance rules require hiding PII and PHI data in audit logs, you can do so by disabling the input and output prompts saving mechanism using the AI Trust Layer policy settings. To do this, take the following steps:

Go to Automation Ops™ > Governance and select the AI Trust Layer policy.
Under the Feature Toggles tab, make sure to set Enable prompts saving for Audit? option to No.

Note:
This configuration allows you to hide sensitive content from log entries, maintain compliance requirements, and control visibility of sensitive data while preserving audit capabilities. However, please note that once hidden, you cannot revover the prompts for further use.

Checking usage summary

The Usage Summary tab on the AI Trust Layer page provides an overview of the model usage and restrictions across different regions. It represents the historical data from your audit log and reflects the settings of your governance policies.

You can view data displayed on the following criteria:

Total LLM Actions per Status: Enables you to monitor the status of different models across regions. To customize the data vizualization, you can filter by region, model, and status.
Total LLM Actions per Product: Allows you to monitor the AI feature adoption within your organization. To customize the data visualization, you can filter by tenant and product.