Microsoft described Copilot as "entertainment only" in terms of use

Microsoft included a clause in Copilot's terms of service that surprised many: the AI assistant is described as intended "only for entertainment." But AI skeptics are not the only ones warning about the risks of blind trust in models. The companies themselves do this in their user agreements, albeit in small print.

Microsoft is far from alone in this market. OpenAI warns in its ToS about ChatGPT's known tendency toward "hallucinations"—confident statements of false facts. Google explicitly states that Gemini's responses should not be viewed as medical, legal, or financial advice.

Meta includes similar liability disclaimers for all of its language models. All these caveats serve one purpose: to legally protect the developer in case a user follows erroneous AI advice and suffers losses. Trusted a chatbot's medical advice—read the ToS.

Made a business decision based on incorrect data—your risk. The paradox lies in this: the very same companies simultaneously aggressively promote their AI tools as reliable work assistants. Microsoft offers corporate clients Microsoft 365 Copilot for $30 per month per user, emphasizing automation, data analysis, and productivity gains.

Marketing materials depict AI as an indispensable tool in medicine, law, finance, and strategic planning. But in the user agreements—a completely different picture. Marketing promises a reliable assistant; the legal department—"only entertainment."

This contradiction is particularly dangerous in the context of how modern language models work. AI systems often respond with high confidence even when they are wrong: models don't say "I don't know," they generate plausible-sounding text that creates an illusion of authenticity. Most users don't read ToS—and a person using Copilot for legal contract analysis or tax risk calculation may simply not realize that the manufacturer itself legally describes the product as entertainment.

The situation is unfolding against a backdrop of intensifying regulatory pressure on the entire industry. The European AI Act is gradually introducing strict requirements for transparency and developer accountability in high-risk systems. In the US, the FTC and relevant Congressional committees are increasingly asking whether companies are misleading consumers by overstating the reliability of their AI products.

Protective clauses in ToS are apparently part of legal preparation for these inspections and potential lawsuits. The phrasing "only for entertainment" in Copilot's terms of service is neither a random detail nor grounds for irony. It is a symptom of a structural gap that the entire industry has not yet overcome: between how AI is sold and what its actual capabilities and limitations are.

While marketing paints one picture, the legal department creates a fundamentally different one. Ultimately, the user must decide for themselves whom to believe—the advertising promise or the fine print of the user agreement.