Teachers unions and Microsoft have agreed on a new nationwide standard for how artificial intelligence tools are used in American classrooms, promising that student and educator data will not be used to train models and that human oversight will be required for any AI decision that affects a student.
The American Federation of Teachers, its New York City affiliate the United Federation of Teachers, and Microsoft announced the National AI Safety and Privacy Standard on September 8. The standard was negotiated over several months in the absence of federal or state rules governing AI in schools. It sets three foundational commitments — protecting children’s privacy and data, giving schools real control over the AI tools used in classrooms, and giving parents plain-language information about how those tools work.
Under the standard, districts can incorporate the protections directly into their existing Microsoft customer agreements, which makes each protection contractually enforceable. Districts that adopt the language can end their agreement and seek damages if a vendor violates it. Microsoft has said it intends to extend the terms to every U.S. school district that wants them.
Three specific requirements sit at the center of the agreement. First, student and educator data cannot be used to train AI models, sold, or repurposed, and schools retain control over how their data is stored, retained, and deleted. Second, AI systems used with students must include human oversight for any consequential decision, and cannot be designed in ways that manipulate students or expose sensitive information. Third, families are entitled to clear, accessible information about how AI is being used in their children’s classrooms and what data those tools collect.
The standard’s text specifies that the data covered by the agreement includes not only student records but the prompts a student types, the AI-generated responses they receive, and the files they upload. Behavioral tracking, long-term profiling, keystroke logging, and continuous attention monitoring are all restricted. Providers are required to notify a school district within 72 hours of becoming aware of a confirmed or reasonably suspected breach involving covered data, and they generally have 180 days after a district’s deletion request to remove the data from active systems, subject to legal holds and district retention policies.
The agreement lands in a policy vacuum. Existing federal privacy laws that apply to education — the Family Educational Rights and Privacy Act, which covers student records, and the Health Insurance Portability and Accountability Act, which covers health information — were written well before generative AI tools were classroom fixtures, and neither addresses how those tools may train on student prompts or generate outputs that affect a student’s grade or placement. That gap has already pushed some states to act on their own: Ohio’s Department of Education and Workforce released a model AI academic policy in January and has since required every public, community, and STEM school in the state to adopt an AI framework by the start of the 2026–27 school year.
The American Federation of Teachers represents about 1.8 million pre-K-through-12th-grade educators, and its New York City affiliate the United Federation of Teachers represents roughly 200,000 members. The standard follows a New York City Public Schools framework adopted the previous week under Mayor Zohran Mamdani that restricted screens and student-facing AI in the youngest grades. Similar frameworks have been picked up by New York State United Teachers and by the Los Angeles Unified School District.
For now the agreement covers Microsoft products only, but the structure — enforceable protections written into an existing vendor contract — is the same one districts already use for privacy language on other software. That gives school boards a familiar mechanism to adopt, and it lets any district that wants stronger protections do so without waiting on Congress or a state legislature.
