Microsoft Called Copilot “For Entertainment Purposes Only”

When we published this piece in April 2026, Microsoft’s Terms of Use for Copilot contained a sentence worth reading twice:

Copilot is for entertainment purposes only. It can make mistakes, and it may not work as intended. Don’t rely on Copilot for important advice. Use Copilot at your own risk.

It was no draft someone forgot to clean up. The terms archive shows the language across several versions. The same company running full-page ads about “AI-powered productivity” had its lawyers classify the product as entertainment.

One Hacker News commenter found the right analogy: “The new ‘for tobacco use only’ of tech.”

Update, October 2026: the version effective 18 August 2026 drops the sentence. It now says: “Do not use Copilot as a substitute for professional advice. Always verify the accuracy of information presented by Copilot before you rely on it.” It still disclaims every warranty, and it states that it doesn’t apply to anyone who signs in with a work, school or organizational (Entra) account, which is how businesses use Microsoft 365 Copilot.

What Microsoft Charges for Copilot

Meanwhile, Microsoft’s pricing for Copilot has become a landscape unto itself. As of October 2026:

Tier	Price	What You Get
Copilot Chat	Free	Web-grounded chat, included with eligible Microsoft 365 plans
Copilot Pro	No longer sold	Replaced by Microsoft 365 Premium
Copilot Business	$21/month, $18 in year one for annual subscriptions bought by 31 December 2026	SMBs up to 300 users
Microsoft 365 Copilot	$30/month	Enterprise organizations
Copilot Studio	$200 per credit pack, or pay-as-you-go	Custom agent building

Copilot Business and Microsoft 365 Copilot each require “a separate license for a qualifying Microsoft 365 plan”, according to Microsoft’s pricing page. For your budget, the Copilot price is only part of the bill.

And Copilot’s terms still can’t promise that its responses won’t infringe someone’s copyright.

The Marketing-Legal Paradox

On Microsoft’s Copilot enterprise page, bold claims: “100%+ Projected ROI with potential payback in 10 months.” Below that: “8+ hours of projected time savings per user per month.”

On the Terms of Use, in all caps:

WE DO NOT MAKE ANY WARRANTY OR REPRESENTATION OF ANY KIND ABOUT COPILOT.

Both pages belong to Microsoft, and both sell the Copilot name. The terms cover the personal Copilot app, while the ROI page sells Microsoft 365 Copilot to businesses. Only one of the two is a legal document.

As Hacker News user andy81 put it: “The only thing clear about that License agreement is it contradicts all their other marketing about Copilot. So either that document is fraudulent or everyone else at Microsoft is committing fraud daily.”

It goes deeper. Earlier this year, Microsoft tried to remove AI disclaimers from Copilot responses in M365, citing user “distraction.” Five days and one backlash later, they reversed course. The disclaimer “will always be visible,” they said — quietly admitting the warnings are load-bearing, not cosmetic.

Why “Entertainment” Was Exactly the Point

It wasn’t sloppy drafting. It was legal architecture.

By classifying Copilot as entertainment, Microsoft insulated itself from product liability claims. If Copilot hallucinates a financial figure in your board presentation, generates content that infringes someone’s copyright, or gives your employee dangerously wrong advice, that’s your problem. The April 2026 terms told you it was entertainment. The current ones say Microsoft is “not responsible for any consequences that arise from your use of or reliance on Copilot.”

Security researcher Reed Mideke nailed it: “Microsoft has no idea how to stop prompt injection or hallucinations, which makes it fundamentally unfit for almost anything serious. The solution? Shift liability to the user.”

For a company buying AI, that makes one question decisive: who carries the risk when the answer is wrong? The marketing sells Copilot as essential infrastructure, and the terms put every consequence on the user. Before a rollout, check which terms your licence falls under and what they promise about accuracy.

96.7% of Users Said No

The market voted. As of early 2026, only 3.3% of Microsoft 365 users paid for Copilot: 15 million seats out of more than 450 million. Of those who paid, barely a third activated it. Of organizations that ran pilots, only 5% moved to broader deployment.

Read those numbers again: 96.7% of the M365 installed base looked at Copilot and said no.

Microsoft’s response looked like panic disguised as strategy. They cut the business price to $18 in a promotion, and have since run a new first-year price of $18 for annual subscriptions bought between 1 July and 31 December 2026. They shoved Copilot buttons into Notepad, Paint, and File Explorer — until users complained and the features were walked back. Microsoft’s share price dropped 15% in early 2026 as investors grew “weary of massive AI capital expenditures that have yet to yield a proportional explosion in revenue.”

As we wrote when the price cut happened: cheaper Copilot still won’t fix your AI problem. The issue was never the price tag. It’s what’s behind it.

Copilot Cowork: The $99 Admission

When your own AI was legally classified as entertainment, what did you do? You bought someone else’s. Twice.

Remember: in October 2025, Microsoft held 27% of OpenAI, a $135 billion stake built on $13 billion in investment. Copilot was built on OpenAI’s models, and Microsoft was OpenAI’s largest external shareholder. And yet, when it came time to build AI that actually does things instead of generating text, they didn’t go to OpenAI. They went to Anthropic.

In March 2026, Microsoft launched Copilot Cowork, their entry into agentic AI. Microsoft didn’t build the agent technology, and they didn’t use their $13 billion OpenAI investment either. They licensed it from Anthropic, a completely separate company, reportedly at roughly $500 million per year.

Microsoft’s own blog was remarkably candid: “Working closely with Anthropic, we have integrated the technology behind Claude Cowork into Microsoft 365 Copilot.” Read that again — the technology behind Claude Cowork. Not inspired by. Not similar to. The actual technology.

At launch, Microsoft put Cowork in the new M365 E7 “Frontier Worker Suite” at $99 per user per month, a 65% jump from E5, and positioned it as a discount on buying the components separately (E5 at $60 + Entra Suite at $12 + Copilot at $30 + Agent 365 at $15 = $117). Today Cowork is billed by usage: “your organization is charged based on how much your users do with Cowork.” The same FAQ confirms that Cowork uses Anthropic’s Claude models as a subprocessor.

Follow the sequence: invest $13 billion in one AI company, build a chatbot on their tech, classify it as entertainment, watch adoption crater, then outsource the hard part to a different AI company and charge for the result. That’s not a product roadmap. That’s two white flags.

The Architecture Gap

The entertainment disclaimer and the Cowork outsourcing pointed to the same root cause: a generational architecture gap.

When Microsoft rushed to ship Copilot in 2023, they bolted their $13 billion OpenAI investment onto Office and Teams: a text-generation sidebar that didn’t remember your last conversation, couldn’t coordinate across tools and took no action on your behalf. Microsoft has added to it since. Copilot’s memory is in preview and stored per user, in that person’s mailbox, and Cowork runs scheduled and email-triggered tasks.

When they realized a chatbot wasn’t enough, they didn’t go back to OpenAI, their own investment. They went to Anthropic, a separate company, for agentic capabilities. Cowork runs on someone else’s models, layered onto Microsoft’s existing stack and metered on top of the licence.

amaiko launched in July 2025, eight months before Cowork existed. Not a sidebar. A multi-agent system where a main agent delegates to specialists, consolidates your team’s conversations into company knowledge every night and works natively in Microsoft Teams. Morning briefings and Active Inbox come with the seat, with unlimited tokens.

AI is hard, and anyone claiming otherwise is selling you something. We’re not going to pretend we’ve solved it. But there’s a clear difference between a vendor that classified its own product as entertainment and one that builds agents designed to do real work. We lay that out in the structural difference between amaiko and Microsoft Copilot.

And when the AI your company sanctions shares its name with a product its own vendor called entertainment, don’t be surprised that employees start bringing their own.