Microsoft Is Exploring an Always-On, OpenClaw-Style Copilot That Would Read Your Inbox and Calendar Overnight

Microsoft, the company that sells productivity software to hundreds of millions of workers, is studying how to make its assistant do the work before anyone asks. According to a report from The Information, picked up by The Verge and TechCrunch, the company is testing OpenClaw-style autonomous agent capabilities inside Microsoft 365 Copilot — a push to turn a reactive chat assistant into a persistent agent that runs around the clock on a user's behalf.

Omar Shahine, a Microsoft corporate vice president, confirmed to The Information that the company is "exploring the potential of technologies like OpenClaw in an enterprise context." That is a careful phrase: no product, no date, no name. But the shape of the experimentation is specific. An always-on version of Copilot could reportedly monitor a user's Outlook inbox and calendar overnight and serve up a list of suggested tasks each morning — the assistant as a night shift that drafts, triages and queues work for review.

Microsoft is also reportedly exploring role-specific versions of the same idea: agents tailored to marketing, sales or accounting, deliberately siloed from the rest of the enterprise system to "limit the permissions the agent needs." That design choice reads as a direct response to the security problems that have followed OpenClaw, the open-source personal agent framework whose viral adoption earlier this year came with a long tail of prompt-injection, credential and over-permission incidents. Sources told The Information that Microsoft believes it can build "safer" versions of the OpenClaw pattern inside its existing enterprise security and compliance stack.

The contrast with the open-source original is instructive. OpenClaw proved that people want an assistant with memory, initiative and continuous access to their mail, calendar, browser and files — and simultaneously demonstrated how dangerous that access is when granted broadly and locally. Microsoft's reported approach inverts the model: narrow scopes per role, no cross-system access, and an enterprise governance layer on top. It is the same idea wearing a compliance uniform.

It would also be a significant escalation for Copilot itself. Microsoft has spent two years positioning Copilot as an assistant embedded in the flow of work; an autonomous version that reads mail and calendars unprompted moves it into the same category as Perplexity's always-on Teammates and the personal-agent products Meta, OpenAI and the Chinese platforms have been racing to ship. The difference is distribution: Copilot sits inside the software most of the corporate world already pays for.

For now, all of this is exploration, not announcement. Microsoft has not confirmed a product, and The Information's report is based on sources describing internal tests. But the direction of travel is clear enough: the next phase of the assistant wars is not about who answers questions best, it is about who is allowed to act — and on whose infrastructure, with whose permissions. The largest enterprise software company on earth just put that question on its own roadmap.