2026 年 5 月

May 21, 2026

Introducing centralized guardrails in public preview

You can now configure organization-wide AI safety guardrails directly from the AI Trust Layer policy in Automation Ops. Use the Centralized guardrails tab to enable validations per product. Once enabled, each guardrail applies automatically to all LLM calls for the selected products across the deployed tenant, without requiring individual agents or automations to be reconfigured.

The following guardrails are now available in public preview:

• Harmful content – Detects hate speech, self-harm, sexual content, and violence in model interactions. Powered by Azure AI Content Safety.
• IP protection – Detects intellectual property leakage in model-generated text and code. Powered by Azure AI Content Safety.
• PII detection – Previously available only at the agent level, now also available as a centralized guardrail. Identifies and blocks sensitive information. Powered by Microsoft Azure Cognitive Service.
• Prompt injection – Previously available only at the agent level, now also available as a centralized guardrail. Detects jailbreak and prompt injection attempts that try to bypass system instructions. Powered by Azure AI Content Safety.

For details, refer to Centralized guardrails.

May 15, 2026

Human-in-the-loop enforcement policy for agent design

The human-in-the-loop enforcement policy is now available in Automation Ops. Administrators can configure this policy to require human oversight checkpoints in agents designed in Studio Web.

When enabled, the policy prevents agents from being created or published without defined human validation or approval points, helping organizations support governance requirements for sensitive, high-impact, or regulated tasks. For more information, refer to Settings for Studio Web Policies.