Microsoft sells Copilot as a work AI, but its terms call it entertainment

Microsoft has once again had to explain what exactly it is selling under the Copilot brand. On one hand, the company is building the assistant into nearly all of its key products and pitching it as a useful work tool. On the other, Copilot's public terms of use state plainly that the service is intended for entertainment only, can make mistakes, and should not be relied on for important advice. For a product Microsoft promotes as a daily AI assistant, that is a harsh and awkward warning.

The disputed wording appears in the Terms of Use for Copilot, which took effect on October 24, 2025. In the warnings section, Microsoft writes that Copilot may not work as intended, that the company gives no guarantees about how it performs, and that users are themselves responsible for the consequences of their actions and for publishing its output. The terms also state separately that they do not apply to Microsoft 365 Copilot unless a specific service explicitly refers to them.

But for an ordinary user, the difference between consumer Copilot, Copilot in Windows and enterprise Copilot is not obvious: the brand is the same, the interface looks similar, and the promise is roughly the same everywhere — speed up work and take away routine tasks. That is why the story drew such a strong reaction.

For several years, Microsoft has been building an image of Copilot as an indispensable AI colleague: the assistant is supposed to help write emails in Outlook, build presentations in PowerPoint, analyze spreadsheets in Excel and find relevant context across work documents. On its official pages, the company talks about higher productivity, saved time and a new model of everyday work with documents and data. Microsoft 365 Copilot costs $30 per user per month with annual billing on Microsoft's website. Next to that price and that marketing, a phrase about "entertainment purposes" in the public rules creates a contradiction too obvious to ignore.

From a legal standpoint, the point of the clause is clear: Microsoft is protecting itself against claims if Copilot produces an incorrect, harmful or controversial result. The company also warns separately that it does not promise that the model's answers will be free of copyright, trademark or privacy violations, and that the user bears responsibility for any public use of the content.

Adoption figures add to the context. In January 2026, Microsoft reported 15 million paid Microsoft 365 Copilot seats. Set against roughly 450 million commercial Microsoft 365 subscriptions, that is about 3.3% of the base. There is growth, but it is still modest for a product that has absorbed tens of billions of dollars and is built into almost every part of the Microsoft ecosystem. Industry research has also pointed to trust problems with Copilot's answers, and to the fact that, given a choice, many employees prefer ChatGPT or Gemini.

After the story spread across social media and trade outlets in early April 2026, Microsoft told journalists that the disputed phrase is outdated text left over from the Bing Chat era and will be updated in the next revision of the terms. But as of April 28, 2026, the wording is still visible on the public Terms of Use page, which is why it keeps working against the company.

The main takeaway is simple: generative AI is already sold as part of office infrastructure, but vendors' level of accountability does not yet match the scale of their promises. While companies keep protecting themselves with language about possible errors and risks, AI assistants remain not a replacement for human judgment but a tool that requires mandatory double-checking. For Microsoft, this is not just a reputational slip but a reminder that the next phase of AI competition will revolve around trust, not just the number of integrations and the loudness of presentations.