Districts can now bind Microsoft to AI privacy terms starting in November

Microsoft is offering US school districts a contract that legally binds it to ten AI privacy principles: no model training on student or educator data, a ban on AI companions, plain-language disclosure to families, and mandatory human review for "high-risk" decisions. Any district can opt in starting in November. The deal lands one week after New York City and Los Angeles each announced one-year bans on student-facing AI tools, the two largest US school systems each choosing to step back from the technology until guardrails are set (The Verge).

AFT President Randi Weingarten called the terms "iron-clad" and warned that "anything less than legally enforceable provisions is simply a wish list," adding that "no one else, including the federal government, has stepped up to do the real work."

A federal AI rule for K-12 does not exist. The US Department of Education has issued nonbinding guidance on student data, and the Federal Trade Commission has pursued student-privacy cases on a complaint basis, but no statute governs how AI vendors handle K-12 data. New York and Los Angeles reached for the lever they had: blocking the tools. Microsoft's contract is a different lever. Districts that opt in buy enforceable obligations in exchange for adoption, and the contract acts as a regulatory substitute until Congress or the Department of Education writes a real rule.

The ten principles cover the points school boards have pressed vendors on for two years. Microsoft commits not to train or fine-tune AI models on student or educator data, to minimize the data its tools collect, to disclose in plain language what is being collected, and to prohibit "AI companions," the chatbots that role-play a friend, tutor, or character for a student. A separate clause requires a human reviewer before any "high-risk" decision, defined in the agreement to include discipline, grading, special-education placement, and recommendations that materially affect a student's record.

Most vendor defaults allow customer data to be used to improve models unless the customer affirmatively opts out. Microsoft's contract reverses that for any district that signs on: no training on student or educator data is the default, and the commitment survives across model versions.

The "high-risk" decisions carve-out is where the agreement meets reality. Discipline and grading decisions are already contested on grounds of bias and due process; layering an AI tool into that process, even with a human reviewer named in the contract, still leaves a district defending its process in a hearing. Special-education placement is the more obvious pressure point because the Individuals with Disabilities Education Act requires specific procedural steps, and an AI tool that informs a placement decision will be subject to the same challenges an evaluation team is.

The contract binds Microsoft, not OpenAI, Google, or Anthropic. Districts running multiple vendors will need separate commitments from each, and a board that treats the Microsoft template as a floor will have to ask the same questions of every other vendor on its roster.

A separate $23 million AI training hub for educators landed the same day, funded jointly by Anthropic, Microsoft, and OpenAI. It is a different program from the privacy principles, with its own curriculum and delivery structure. A district that opts into the privacy contract is not signing up for the training hub, and a district that signs onto the training program is not buying the privacy terms.

The agreement is opt-in, and the easiest signal is the school board agenda over the next 60 days. Districts that adopt the template will publish it as part of their vendor agreements; districts that decline will have to explain to families why they chose voluntary policy terms over a binding instrument.

Districts have roughly four months before the November opt-in window opens. The choice is the one Weingarten named: enforceable terms, or what she called a wish list.