Microsoft says computer-using agents in Copilot Studio now offer more model choice, stronger security and governance, and easier scale to automate web and desktop apps across enterprises.
The changes respond to early customer feedback after the initial launch, the company said in an announcement, which highlighted limits around scaling, authentication, and observability.
Microsoft added support for multiple foundation models, including Anthropic’s Claude Sonnet 4.5 and OpenAI’s Computer‑Using Agent, letting teams choose models for dynamic interfaces or multistep orchestration. In the announcement, the big tech firm noted that built‑in credentials enable unattended runs and can be stored either internally (encrypted in Microsoft Power Platform) or in Azure Key Vault; credentials are encrypted and are never exposed to the AI model.
Related reading
- Nadella looks beyond the AI hype cycle as industry turns to delivery in 2026
- Wondershare launches EdrawMax V15
- impact.com launches marketplace for content creators to monetise AI
Monitoring and audit capabilities are expanded with session replay and screenshots, step‑by‑step action logs (including coordinates and timestamps), run summaries, resource tracking and export options. Integration with Microsoft Purview, Dataverse logging with configurable verbosity (All data, Data without screenshots, or Minimal), and retention choices from 7 days to indefinite are also included.
Microsoft also offers managed Cloud PC pools powered by Windows 365 for Agents to provide scalable, Microsoft Entra‑joined and Intune‑enrolled capacity. Tenants can create up to two Cloud PC pools for evaluation, with 50 hours of free usage for published autonomous agents, it added.
The recap
Microsoft expands Copilot Studio computer-using agents with security and scale
Supports Anthropic Claude Sonnet 4.5 and OpenAI Computer‑Using Agent models
Two Cloud PC pools per tenant with 50 free hours