Overview
- Microsoft, which on Monday launched Critique and Council inside Copilot’s Researcher, is making both features available in its Microsoft 365 Frontier early-access program.
- Critique pairs one model to draft and another to review for accuracy, completeness, and citation integrity, using Anthropic and OpenAI models with plans to enable either model to take the reviewer role.
- Microsoft reports a +7.0 point improvement on the 100‑task DRACO deep‑research benchmark versus the best prior system, with the largest gains in depth of analysis and presentation quality.
- Council runs Anthropic and OpenAI models side by side, shows full reports from each, then uses a judge model to summarize where the reports align or diverge and what each adds.
- Copilot Cowork, based on Anthropic’s agent, is now in Frontier and runs in tenant‑sandboxed environments so customer data stays under company control, a setup meant to reduce hallucinations and speed real work.