Microsoft Just Promised Not to Train AI on Your Kid's Data

Yesterday, in New York, the two largest teachers’ unions in the country stood next to Microsoft’s president and signed something that’s never existed before: a set of AI rules for schools that a district can actually take to court. The National AI Safety & Privacy Standard, announced September 9, 2026 by the American Federation of Teachers, the United Federation of Teachers, and Microsoft, is the first agreement of its kind. Not a pledge. Not a blog post about “responsible AI.” A standard districts can write directly into their Microsoft contracts, starting November 1.

Here’s why that matters more than it sounds like it should: there is currently no federal law and almost no state law governing what AI companies can do with a student’s data once it enters a school-issued account. Nothing. This deal exists because Congress hasn’t, and because 50 state legislatures haven’t either. Two unions and one company just built the fence themselves.

The short version

If you read nothing else, read this.

What’s actually happening	What it means for your family
AFT, UFT, and Microsoft announced the National AI Safety & Privacy Standard on Sept. 9, 2026	This is the first legally enforceable AI privacy deal for K-12 schools in the country
Microsoft can’t use student or teacher data — prompts, AI-generated work, uploaded files — to train or improve its models, with narrow exceptions for safety and security	Your kid’s homework doesn’t quietly become training material for the next version of Microsoft’s AI
Districts can add the standard to Microsoft contracts starting Nov. 1, 2026, with no renewal or renegotiation required	It’s not automatic. Your district still has to ask for it
Any AI-driven decision affecting a student requires human oversight, and AI can’t be used to track students	A tool can flag something. A person still has to decide what happens to your kid because of it
Only Microsoft has signed so far, though AFT wants OpenAI and Anthropic to follow	This covers Microsoft’s education-specific AI tools. Whether it also covers everyday Copilot — the tool most kids would actually reach for — is unclear, and it doesn’t cover any other company’s AI product in your kid’s classroom
The UFT is keeping its own one-year ban on student AI use in NYC elementary and middle schools in place, even after signing this deal	The union that negotiated this still doesn’t think middle schoolers should be using these tools unsupervised

What does the National AI Safety & Privacy Standard actually protect?

Strip away the announcement language and here’s what the standard actually locks in, in plain terms:

1. No training on student or teacher data. Prompts, AI-generated responses, and uploaded files can’t be used to train or improve Microsoft’s general-purpose AI models, sold, or repurposed for anything unrelated to the school’s use — outside narrow safety and security exceptions.
2. No tracking students. The standard bars using AI systems to monitor or track students, closing off one of the more unsettling directions “personalized learning” software has quietly drifted toward.
3. A human has to be in the loop. Any AI-driven decision that affects a student requires human oversight before it counts. The software can flag, suggest, or draft. A person still has to sign off on anything that touches a kid.
4. Plain-language transparency for parents and teachers. Districts and families get actual explanations of how the tools work, not a 40-page terms-of-service document nobody reads.
5. It’s contractually enforceable, not just promised. Districts write this directly into their Microsoft customer agreement. That turns “we care about student privacy” from a marketing line into a term Microsoft has to answer for if it breaks it.

That fifth point is the one worth sitting with. Companies make privacy promises constantly. Almost none of them are things a school district’s lawyer can point to in a contract dispute. This one is.

Why a corporate handshake is doing a legislature’s job

This site covered AI homework use a couple weeks ago — 86% of kids 9 to 17 are already using generative AI, most of them for schoolwork. What that post didn’t get into is the question underneath all of it: where does everything your kid types into a school AI tool actually go? Nobody had a real answer. That’s the gap this agreement is trying to fill.

AFT President Randi Weingarten didn’t mince words about why a union had to negotiate this instead of a legislature passing it: “Anything less than legally enforceable provisions is simply a wish list.” Microsoft Vice Chair Brad Smith called it a standard meant to set “a high bar for child privacy and AI safety” and committed to extending it to every district in the country. Those are strong words from both sides of the table, and for once the paperwork actually backs them up.

But notice what’s really happening here. Two unions representing teachers negotiated privacy terms that, in most other industries, a regulator would set. That’s not a knock on the unions — it’s the opposite. It’s a sign of how far behind actual policy has fallen. Your kid’s school already treats phones as something to actively manage instead of trusting an app’s settings menu. This is that same instinct, aimed at a bigger and less visible problem, built by the people closest to the classroom because nobody above them got there first.

The parts worth staying skeptical about

I don’t want to hand you the press-release version of this and call it a day, because the honest version has real caveats in it.

It’s opt-in, not automatic. The protections only apply if your district actually adds them to its Microsoft contract. A standard nobody requests doesn’t protect anybody. If your district already uses Microsoft’s education tools, find out — don’t assume — whether it’s signed on.

It’s one company — and maybe not even all of it. Right now this covers Microsoft’s education-specific AI products. Whether it also covers everyday Copilot, the general-purpose assistant most kids and teachers would actually open, is genuinely murky: Microsoft’s own announcement never uses the word “Copilot,” and the standard’s fine print excludes general-purpose productivity, collaboration, and search tools that aren’t “primarily designed” for education — a description that arguably fits the Copilot in front of your kid. Nothing about this deal touches Google, OpenAI, or the dozens of smaller AI tools your kid’s teachers might be piloting this year, either. Fairplay, a children’s advocacy group, pointed out exactly this: the deal “does not fool schools into thinking that GenAI products are safe and effective,” and it “only applies to Microsoft, not the many other companies selling AI products to schools.”

De-identified isn’t the same as anonymous. Even under strong privacy terms, researchers have shown that “anonymized” data can sometimes be pieced back together from writing style, background details, and other small signals. A contract can restrict what a company is allowed to do with data. It can’t make re-identification technically impossible.

It stops at the schoolhouse door. Fairplay’s sharper point is the one that should stick with you: AI companies want kids who use their products all day in class to keep using them at home, “where there are likely to be no protections.” This agreement governs the school account. It says nothing about the same kid, the same app, on a personal device at 9 p.m.

None of that makes the agreement meaningless. It makes it a floor, not a finish line. A real one, for the first time — just not the whole fence.

What this looks like at your kitchen table

Say your seventh grader comes home and mentions the class is using Copilot to help draft a book report. Here’s the honest answer, not the tidy one: it’s not clear that this standard actually covers the Copilot your kid just used. Microsoft’s own announcement never says the word, and the fine print carves out general-purpose tools that aren’t built specifically for education. A year ago, you had no real way to know what happened to that draft after it left the screen — whether it sat somewhere, got used to train the next model, got sold, got anything. Now there’s at least a document to point to, and a specific question to ask your district: does the tool your kid actually used tonight fall inside this standard, or outside it?

That’s the shift. Not that AI in your kid’s classroom suddenly became risk-free — it didn’t, and no contract makes it so. It’s that for the first time, “is this safe” has a real document behind it instead of a shrug. I wish every piece of technology that touched your kids came with that kind of paper trail. This is one of the first that does, and it exists because people who actually stood in front of a classroom decided a wish list wasn’t good enough anymore.

What to do this week

• Ask your district, directly, if it has adopted the National AI Safety & Privacy Standard, or plans to once it’s available November 1. A short email to the superintendent or IT department gets you a real answer.
• Ask what AI tools beyond Microsoft your kid’s school is using. This standard doesn’t cover them. You want to know what’s in the building either way.
• Read whatever plain-language explanation the district sends home about its AI tools. The standard requires one. Actually read it instead of filing it with the rest of the back-to-school paperwork.
• Talk to your kid about the home-use gap. The protections stop at the school account. If they’re using the same chatbot on a personal phone at night, none of this applies — that conversation is still yours to have.
• Don’t treat this as “solved.” The same instinct that had the site checking Meta’s teen settlement for the catch applies here. Read the win. Then read what it doesn’t cover.

The takeaway

For the first time, “we won’t use your kid’s data to train our AI” is a sentence with a contract behind it instead of a sentence in a press release. That’s real progress, and it’s still just one company, one agreement, and one piece of a much bigger gap nobody in Washington has bothered to close yet.