Microsoft builds governed agents

Microsoft is developing Copilot features modeled on the OpenClaw agent approach to give enterprises stronger security controls, role-specific agents, and more governable automation. Separately, Bloomberg reports Microsoft has taken over data-centre capacity in Norway that had been marketed for OpenAI, signalling a move to secure infrastructure as it rolls out those enterprise capabilities. (computerworld.com) (bloomberg.com)

Microsoft is building more autonomous Copilot agents for workplace software while tightening the security controls that large companies demand. (computerworld.com) The new work is modeled on OpenClaw, an open-source tool that can run on a computer and carry out tasks for a user over time instead of waiting for one prompt at a time. Microsoft confirmed to The Information, via reports in Computerworld and TechCrunch on April 13 and April 14, that it is exploring that approach for Microsoft 365 Copilot. (computerworld.com) (techcrunch.com) Microsoft’s version is aimed at enterprise customers, with tighter access controls and role-specific behavior for jobs such as sales, marketing, and accounting. TechCrunch reported the concept includes an “always working” Copilot that can handle multistep tasks over long periods. (techcrunch.com) (computerworld.com) The basic problem Microsoft is trying to solve is simple: companies want agents that can do more than answer questions, but they also need to see, limit, and stop what those agents do. Microsoft’s own Copilot material now frames “trust,” “governance,” and “observable” work as core requirements for agents inside Microsoft 365. (microsoft.com) (adoption.microsoft.com) That push has been building for months. On March 9, Microsoft said Microsoft 365 Copilot Wave 3 moves beyond assistance to “embedded agentic capabilities,” and described Copilot Cowork as long-running work that can be reviewed, guided, or stopped inside Microsoft’s security and identity framework. (microsoft.com) Microsoft has also been telling information technology teams to prepare for agents at scale. In a January 26 Copilot Studio post, the company said organizations want agents that can own workflows end to end, act across systems, and scale “without sacrificing control.” (microsoft.com) The infrastructure side is moving in the same direction. Bloomberg reported on April 14 that Microsoft agreed to rent capacity at a Norway site that had been marketed for OpenAI’s Stargate effort, including 30,000 additional Nvidia Vera Rubin chips at Nscale’s Narvik campus, on top of a prior $6.2 billion commitment there. (bloomberg.com) CNBC reported on April 15 that OpenAI pulled back from renting capacity directly at the Narvik project and is instead discussing renting that compute from Microsoft, while Microsoft said the expansion with Nscale is meant to ensure customers have advanced artificial intelligence infrastructure in Europe. (cnbc.com) Microsoft has not publicly launched the OpenClaw-style Copilot features yet, and reports point to a possible preview around the company’s Build conference in June 2026. For now, the picture is a company lining up both the guardrails and the graphics processors before it asks enterprises to trust agents with more of their work. (techcrunch.com) (blogs.microsoft.com)

Get your own daily briefing

Scout delivers personalized news, insights, and conversations tailored to your role and industry.

Download on the App Store

Shared from Scout - Be the smartest in the room.