Microsoft's Copilot terms of use describe the assistant as "for entertainment purposes only"

Microsoft has put wording in the Copilot terms of use that surprised many people: the AI assistant is described as intended "for entertainment purposes only." But AI skeptics are not the only ones warning about the risks of blindly trusting models. The companies themselves do it in their user agreements, albeit in fine print.

Microsoft is far from an exception in this market. OpenAI's ToS warn of ChatGPT's known tendency to "hallucinate," that is, to state false facts with confidence. Google states outright that Gemini's answers should not be taken as medical, legal or financial advice.

Meta includes similar liability limits for all of its language models. All these caveats serve one purpose: to legally protect the developer if a user follows bad AI advice and suffers losses. Trusted a chatbot's medical advice?

Read the ToS. Made a business decision on incorrect data? Your risk.

The paradox is this: the same companies are simultaneously pushing their AI tools aggressively as dependable work assistants. Microsoft sells corporate customers Microsoft 365 Copilot for $30 per user per month, emphasizing automation, data analysis and productivity gains. Advertising materials portray AI as an indispensable helper in medicine, law, finance and strategic planning.

But the user agreements paint a very different picture. Marketing promises a dependable assistant; the legal department says "entertainment only." This contradiction is especially dangerous given how modern language models work.

AI systems often answer with high confidence even when they are wrong: models do not say "I don't know," they generate plausible-sounding text that creates an illusion of reliability. Most users do not read the ToS, and someone using Copilot to analyze a contract or calculate tax risks may simply not realize that the vendor itself legally describes the product as entertainment. The situation is unfolding amid growing regulatory pressure on the whole industry.

The European AI Act is phasing in strict transparency and accountability requirements for developers of high-risk systems. In the US, the FTC and relevant congressional committees are increasingly asking whether companies mislead consumers by overstating the reliability of their AI products. The protective clauses in the ToS are, apparently, part of legal preparation for these scrutiny efforts and potential lawsuits.

The "for entertainment purposes only" wording in Copilot's terms is not a random detail and not a reason for mockery. It is a symptom of a structural gap the whole industry has yet to close: between how AI is sold and what its real capabilities and limits are. While marketing paints one picture, the legal department builds a fundamentally different one.

In the end, users have to decide for themselves whom to believe: the advertising promise or the fine print of the user agreement.

*Meta is designated an extremist organization and banned in Russia.