Auto Critique and Model Council features in Researcher (Frontier program)

Updated April 29, 2026: We have updated the content. Thank you for your patience.

Introduction

We’re introducing new capabilities in Researcher to help teams review AI‑generated research more confidently and reduce rework before using outputs in business decisions.

Critique is a built-in review step in Researcher that automatically checks and improves a draft before it’s delivered. It uses a second reasoning pass to strengthen structure and completeness, prioritize reputable sources, and ensure key statements are grounded in clear citations—so the final report is more reliable and easier to trust.

Council is Researcher’s multi-model comparison mode. It runs the same question through multiple deep‑reasoning research agents (for example, GPT and Claude) in parallel, preserves each model’s full standalone report, and adds a lightweight synthesis “cover letter” that highlights where the models agree, where they diverge, and what each uniquely contributes—so you can compare perspectives and make decisions with higher confidence.

These preview features are initially available only to organizations enrolled in the Microsoft 365 Copilot Frontier program, which provides early access to experimental Copilot capabilities.

This message is associated with Roadmaps 553213 and 558538.

When this will happen:

Public preview: Rolling out to Frontier customers by March 30, 2026.

How this affects your organization:

Who is affected:

• Organizations enrolled in the Microsoft 365 Copilot Frontier program using Researcher
• Admins managing AI model access, including third-party models such as Anthropic Claude
• Users who generate long-form research reports using Researcher
• A Microsoft 365 Copilot (Premium) license is required to use this feature

What will happen:

• Researcher drafts will run through Critique before delivery, improving structure, completeness, and citation quality, when “Critique” is selected as the option in the model picker. 
• Users will be able to enable Council, which generates:
  • Two independent, full research reports from different reasoning models (for example, GPT and Claude).
  • A concise synthesis “cover letter” summarizing areas of agreement, disagreement, and unique contributions.
• Critique and Council availability is tenant-admin controlled, including the ability to allow or block third-party models such as Claude.
• There is no change to existing user workflows unless the feature is enabled.
• Existing security, compliance, and data handling policies continue to apply.

What you can do to prepare:

• Review your AI model access settings and confirm whether Anthropic Claude is enabled for your tenant.
• Decide whether to allow Critique and Council based on your organization’s governance and compliance requirements. There are no separate controls for these features; both are enabled when Anthropic and Claude report generation are turned on.
• Inform helpdesk and support teams about the upcoming preview features.
• Update internal guidance or training materials if Researcher is documented in your organization.

No action is required if you do not plan to enable the preview features.

Learn more: Use model choice in the Researcher agent | Microsoft Support

Compliance considerations:

No compliance considerations identified, review as appropriate for your organization.