Microsoft Power Platform governance and administration - Safety Grader

Update: Release of this feature has been updated.

We are announcing the ability to evaluate AI responses using configurable safety graders to identify unsafe or policy-violating outputs in Microsoft Power Platform governance and administration. This feature will reach public preview on August 31, 2026.

How does this affect me?
Admins will be able to configure specific safety criteria to evaluate AI responses prior to them being sent to identify policy-violating outputs. Once configured, the Safety Grader will be able to do the following:

• Evaluate responses against safety criteria.
• Detect unsafe or undesirable outputs.
• Combine safety evaluations with other graders.
• Provide detailed explanations and evidence supporting the evaluation for each result.

What action do I need to take?
This message is for awareness, and no action is required.