UiPath opens its AI Trust Layer to customer guardrails in preview

UiPath is giving customers another choice in the control layer around its agents. Its early-October platform release notes describe using an organization’s own guardrail subscription, which can matter when safety services must fit an existing vendor agreement or regional deployment.

The October 15 release summary labels bring-your-own guardrails as a preview and names Azure AI Language, Azure Content Safety and Noma as supported during that phase. It also describes Agent review, in preview, replacing the previous Agent score and allowing Autopilot to identify and help fix issues.

A review grade is not an operating guarantee

Changing the review interface can improve feedback during development, but a grade does not establish that an agent will handle every real input correctly. The release also says the former minimum-agent-score publishing policy is no longer enforced because Agent review replaces that score. Administrators should account for that policy change rather than assume the old gate remains active.

The practical configuration question is which guardrails execute for which agent, where their requests are processed and how a rejected or unavailable guardrail affects the workflow. The preview status means those details deserve a controlled rollout and supported configuration review.

Test the configured policy, not a marketing label

Customers can begin with permitted and prohibited examples from the actual job, then inspect both the guardrail response and any downstream action. An unavailable check should not be mistaken for an approving check. Those tests must be repeated after changing providers or policy.

UiPath’s update could align agent controls more closely with enterprise requirements. The next evidence is consistent handling of allowed, blocked and failed checks in an operational workflow, with the review history retained.