Microsoft Admitted On Its Own Terms That Copilot Is Just Entertainment — And The Numbers Prove It

"Copilot is for entertainment purposes only. It may make mistakes and may not work as intended. Do not rely on Copilot for important advice. Use Copilot at your own risk."

— Microsoft Copilot official Terms of Use, IMPORTANT DISCLOSURES & WARNINGS section, updated October 24, 2025.

It's not a paraphrase. It's not a journalist's critique. It's what Microsoft itself wrote, in capital letters, on the official terms page of the product the company promotes, integrates, and charges for the most.

For months, no one read it — because no one reads terms of use. Until, in early April 2026, a Reddit user found the passage and posted it. What followed was the most complete corporate contradiction the AI market has ever seen.

Two Copilots, Two Discourses

To understand the controversy, it's necessary to separate what Microsoft sells as Copilot — because there's more than one.

The controversial terms apply to the consumer Copilot — the one in Windows 11, on copilot.com, on mobile, in other Microsoft apps and websites. The enterprise Microsoft 365 Copilot is explicitly out of scope.

The Chasm Between What Microsoft Sells and What It Delivers

This is the central paradox: Microsoft sells productivity transformation with one hand and transfers all risk to the user with the other — and kept both narratives coexisting in the same product for months without anyone noticing.

The Numbers That Explain Why This Happened

Copilot is not performing well commercially. And the 2026 data is embarrassing:

There's more: when users were asked about preference between Copilot, ChatGPT, and Gemini, only 8% chose Microsoft's product. The NPS (Net Promoter Score) — a satisfaction and recommendation metric — went from -3.5 in July 2025 to -24.1 in September 2025. A negative score means there are more detractors than promoters.

The Detail No One Expected: Microsoft Recommended Using Other Tools

If the terms of use were already embarrassing, an internal leak added another layer: Microsoft itself even internally recommended that its developers use other AI tools instead of GitHub Copilot for some programming activities.

The distance between the external discourse and the internal reality became explicit: the company that talks most about productivity revolution via Copilot is the same one that, behind the scenes, pointed out the product's limitations to its own teams.

Microsoft's Defense — And What It Reveals

When the controversy went viral, Microsoft gave a response to Windows Latest and Fast Company:

"The phrase 'for entertainment purposes' is legacy language from when Copilot was originally launched as a search assistant in Bing. As the product has evolved, this language no longer reflects how Copilot is used today and will be updated in the next revision."

The explanation is plausible. The problem is what it implies: the terms were updated on October 24, 2025 — more than two years after Copilot stopped being Bing Chat and became a central product in the company's strategy. No one in legal realized that the product had changed radically, but the language remained the same.

The Real Problem Behind the Joke: Automation Bias

As ironic as the term "entertainment" may be, the legal warning points to something real: language models hallucinate. They produce fluent and seemingly coherent responses about things that didn't happen, people who don't exist, laws that don't exist.

The concrete risk is automation bias: humans tend to accept outputs from automated systems without questioning, even when the data indicates an error. The more fluent and confident the answer sounds, the harder it is to identify when it's wrong.

In January 2026, Copilot generated false claims about football-related violence — a new episode in a growing list of public product failures. The legal warning exists for a concrete reason, not just for lawyers' caution.

What Microsoft Is Doing About It

Satya Nadella's response was to personally take control of AI product development in September 2025 — a clear sign that Copilot's results were not meeting expectations.

In April 2026, Microsoft launched its first proprietary models: MAI-Transcribe-1, MAI-Voice-1, and MAI-Image-2 — the first since renegotiating the contract with OpenAI in September 2025. This signals a strategy to reduce dependence on the external models that currently underpin Copilot.

What This Changes for You in Practice

If you use the consumer Copilot for real professional tasks — code review, document analysis, decision support — the practical lesson hasn't changed with the controversy:

Always verify. Any LLM can hallucinate at any time. The fluency of the answer is not correlated with its accuracy. Microsoft has merely made this official in legal language.

The real news is different: now you know that if a Copilot error causes a problem, Microsoft has already protected itself. The burden is entirely yours.

Copilot is not a bad product because it's classified as entertainment. It's a problematic product because the company sells one thing and assumes responsibility for another.

This is not exclusive to Microsoft. It is the business model of the entire generative AI industry — charging for the promise, transferring the risk for the outcome.

Microsoft was merely honest enough to put it in writing.