New M365 Copilot Critique and Council features

M365 Copilot Researcher agent is adding new capabilities for deep, complex research. There are two new features: Critique and Council.

Critique is a multi-model research workflow that lets you use different models including to research.  This allows you to use both Anthropic and OpenAI to generate and improve an answer.  For instance, one model plans the work, retrieves information, and drafts the response; a second model reviews and refines the draft to produce the final output. Then the second model focuses on review and refinement, acting as an expert reviewer of the final research response. 

Council is designed for side-by-side model comparisons. Council is different approach and is designed for side-by-side comparison across multiple models.  In this case, Councill will run both Anthropic and OpenAI in parallel and will present a report shown at the end as a comparison between both models.  Once this is completed, a dedicated judge model evaluates both reports and then uses a judge model to synthesize the key findings.

Reference: https://techcommunity.microsoft.com/blog/microsoft365copilotblog/introducing-multi-model-intelligence-in-researcher/4506011

New M365 Copilot DLP Capabilities

Microsoft 365 Copilot now includes new data loss prevention (DLP) capabilities customers should be aware of. Until now, I have typically recommended using SharePoint Advanced Management (SAM) to reduce oversharing and Sensitivity Labels to prevent Copilot from processing sensitive documents even when users have access. There are some new M365 Copilot DLP capabilities that customers should know about.  Today, there are some really good things customers can figure with M365 Copilot to protect against data loss.  Up to now, I have usually talked with customers about implementing SharePoint Advanced Management (SAM) to restrict oversharing of data plus I have recommended implementing Sensitivity Label policies to restrict sensitive documents from being processed by M365 Copilot (even if the user has legitimate access to the data).

A new preview feature lets you apply DLP policies to Copilot prompts themselves, blocking prompts that contain configured sensitive information types. This real-There is a new feature that is in preview that enables you to create DLP policy to protect sensitive information types from even being submitted in M365 Copilot Prompts.  This real time control helps reduce data leakage and oversharing. You can also   Users would not be allowed to even allow prompts to continue if there is sensitive information present.  You can also create policy that would allow users to work with sensitive content but restrict where the prompt can be sent for example, limiting it to internal processing only.

References:

https://learn.microsoft.com/en-us/purview/ai-m365-copilot 

https://learn.microsoft.com/en-us/purview/dlp-microsoft365-copilot-location-learn-about

https://www.microsoft.com/en-us/microsoft-365/roadmap?searchterms=515945

https://www.microsoft.com/en-us/microsoft-365/roadmap?filters=&searchterms=548671

https://techcommunity.microsoft.com/blog/microsoft365copilotblog/what%E2%80%99s-new-in-microsoft-365-copilot--march-2026/4506322

https://techcommunity.microsoft.com/blog/microsoft365copilotblog/latest-enhancements-for-copilot-security-management-and-analytics/4508476