AI Governance - July 17, 2026 - 202607.1.0

Protect critical AI systems by introducing AI Guard SDK by OneTrust, a comprehensive classification and inspection layer that helps ensure AI interactions are safe, secure, and compliant during development and runtime. This enhancement helps identify PII risks early through advanced data classification, enabling teams to test AI agents and quantify PII risk before moving to production. Integrate easily using the SDK-based approach designed for Python runtime environments, ideal for build-phase validation rather than high-volume, externally facing AI use cases.

For more information, see OneTrust AI Guard SDK Guide.

Extend continuous AI monitoring for Amazon Bedrock to gain visibility into AI agent and model activity. This enhancement ongoing governance by capturing agent and model metadata, usage information, evaluation metrics and more directly from Amazon Bedrock powered AI systems as they operate in production.

Additionally, you can detect and measure PII leakage for Amazon Bedrock using the AI Log Analyzer by scanning AI logs stored in Amazon S3, identifying sensitive prompts and responses, and surfacing PII counts in OneTrust for the associated AI assets to reduce risk from non-deterministic AI behavior.

For more information, see the following:

• Observability Usage of AI Governance Agents & Models

• AI Governance Model Evaluation Metrics

• AI Policy Enforcement Guide

• OneTrust AI Log Analyzer

Enable continuous monitoring for AI agents and models built on Microsoft Foundry by surfacing Evaluation Metrics for Agents and Models shifting from AI systems. This enhancement helps shift from one-time assessments to ongoing evaluation using live telemetry from Foundry-based AI systems.

For more information, see the following:

• AI Governance Model Evaluation Metrics

• AI Policy Enforcement Guide

Automatically detect AI policy violations across platforms to provide AI governance teams with actionable steps to ensure governance and compliance throughout an AI Agent lifecycle. This policy orchestration experience allows users to create custom policies for unique scenarios and support guardrail enforcement and detection.

To compliment AI policy management, seeded AI policies are also available to flag and provide enforceable steps on guardrails. This reduces the time to value for customers getting started with AI Policies and provides a template for them to use when defining their own custom AI Policies in OneTrust.

For more information, see AI Policies Feature Guide.

Automatically enforce responsible AI policies on violating agents and models across AI development platforms. This enhancement to AI policies provides consistent guardrail visibility and enforcement measures in the OneTrust Platform.

For more information, see AI Policy Enforcement Guide.

Send customized, trigger-based automated email messages from a new global Notifications tab on the Assessment templates screen. This enhancement speeds up assessment completion, reduces support requirements, and provides users with more relevant information in the assessment.

For more information, see Configuring Custom Email Notifications on Assessment Templates. This feature is related to the following ideas: OT-I-949, OT-I-1270, OT-I-3522, OT-I-7945, and OT-I-16348.

Configure approver-only sections on assessments automatically using template rules or inventory-based values. These enhancements mean users can launch assessments with approver-only sections directly from the self-service portal.

For more information, see Configuring Approver-only Assessment Sections. This feature is related to the following ideas: AA-I-293 and AA-I-302.

Opt-out of automatically scanning your three most recent assessments, plus add more files, more file types, and bigger files to your uploads. These enhancements improve the performance and accuracy of the AI Doc Scanner.

For more information, see Leveraging AI Auto-responses for Assessments.

Issues that may impact our customers are documented in our Known Defects Database located here with approved workarounds and targeted releases for resolving the issue.