Schools get a contract playbook for AI data and oversight

Microsoft, the American Federation of Teachers and the United Federation of Teachers have published a contract standard that gives U.S. school districts a route to demand tighter limits on how AI providers handle student and educator data.

The National AI Safety & Privacy Standard, announced on September 9, is a binding memorandum between the National Academy for AI Instruction and participating AI providers. It says districts can request its protections in their own agreements for educational AI products. The document says providers should make that process available within 90 days of its effective date, although it does not name participating providers beyond the announced Microsoft arrangement.

The standard’s core promise is narrow but consequential: covered student and educator data cannot be used to train, fine-tune or update an AI model, except for a tightly limited safety-and-security purpose. It also requires providers to give customers control over retention and deletion, and to make persistent memory optional.

Its terms reach beyond data use. The agreement requires meaningful human review for decisions affecting educators and students, bans automated grading without review and blocks automated discipline, placement and emotional assessment. It says action-taking features for students must be disabled by default unless a school administrator enables them with scoped permissions, oversight and audit logging.

The privacy protections are not a new federal rule, and they do not automatically bind every school or AI vendor. The memorandum applies to providers engaged by the Academy; other districts would need to request comparable terms in their own contracts. The agreement also allows protections to be adapted into an existing data-privacy or licensing agreement rather than copied word for word.

That contract-by-contract structure is the important limit. Schools that want the standard’s protections still need to procure them, and the document leaves implementation questions to individual agreements. But it turns several commonly stated principles—no training on student data, human review, breach notification and independent certifications—into terms a customer can seek to enforce.

For districts, the immediate test is practical: whether the promised request process is clear, and whether the protections survive the details of product scope, telemetry, sub-processors and enforcement. The standard defines de-identified telemetry separately, permits it for operational uses, and bars its use for generative-model training, profiling or advertising.