Note
Access to this page requires authorization. You can try signing in or changing directories.
Access to this page requires authorization. You can try changing directories.
This article describes the fixes and improvements to existing functionality that are included in this build.
Visit the Microsoft Copilot Studio Blog to stay up-to-date on the latest news!
Version 2026.5.5 updates
New Features
- External agents using the Agent-to-Agent protocol can retrieve current task state via the
tasks/getoperation. - Agents using the GPT-55-Chat model series can run inline agent execution for more capable tool-based orchestration.
- Admins and makers can view environment-level average agent quality and health metrics, with paging controls for the agents inventory.
- Maker evaluation scenarios can set client-side content moderation levels per prompt instead of always using the default High setting.
- Puerto Rican Spanish (es-PR) language support is now generally available.
- Enhanced analytics insights: reaction and feedback for enhanced task completion sessions, plus the ability to show user sentiment and connected agents in summaries.
Fixes and Improvements
- Breaking changes: Custom metrics analytics summaries will always include label trend and over-time fields (even when empty), which may affect integrations that expected these fields to be missing or null.
- Search and snippet relevance improvements: prioritize viewable SharePoint and OneDrive content, prevent universal search failures from duplicate knowledge sources, and rely on native handling for encrypted files; snippets will not be returned from files disallowed for content extraction.
- Connector and action robustness: fewer invalid/undefined schema failures, correct handling of arrays and loosely typed JSON, resilience to duplicate parameter names, reduced permission errors with maker-shared connectors, and more reliable On-Behalf-Of authentication in invoker mode.
- Microsoft Teams and voice calling improvements: 1:1 calling greetings play on callbacks with correct speech markup and recognition, connected agents in group chats/channels respond on first message, and DTMF-only voice agents end calls cleanly when grammar is invalid.
- Publishing, governance, and interoperability: more reliable recurrence trigger publishing with correct UTC/time zone handling, improved resilience during Dataverse outages, automatic recovery after environment copy via identity rotation (with some settings requiring reconfiguration), teammate mode publishing blocked when TaskDialog/AdaptiveDialog components are present, and sharing links for unpublished agents fixed.
- Continued improvements to performance and reliability.
Security
- Error details from Anthropic responses are redacted to prevent customer data from appearing in logs.
- External agent outputs can be marked sensitive based on the agent's configured sensitivity level to prevent forwarding sensitive messages to other external agents.
- Admin controls to suppress or neutralize AI-generated URLs are enabled by default to enforce URL governance more consistently.
- Agent Status displays more detailed Data Loss Prevention messages, categorizing violation details such as connectors and HTTP requests.
- Information Barriers are enforced: agents using Embedded Knowledge cannot be published or shared beyond "Only Me" in tenants with Information Barriers enabled.
- Search and snippet relevance respects SharePoint and OneDrive files that disallow content extraction, preventing snippets from those files.
Return to the Microsoft Copilot Studio Release Notes.