Microsoft hands school districts the right to cancel and sue over classroom AI breaches

The term that decides whether any of this is worth anything is the one that turns a protection into a remedy. Microsoft's obligations switch on for every district it works with on November 1 whether or not the district lifts a finger ^[7], so the no-training and no-tracking commitments are unilateral from that date ^[4]; the right to terminate and claim damages, though, exists only where a district has written the language into its own Microsoft agreement ^[3][5]. A district that does nothing gets a promise its superintendent can repeat to a parent. Spend a lawyer's afternoon on it instead, and a district gets a cancellation right against the largest software vendor in the building.

Between Fortune's September 9 report and the effective date there are 53 days ^[17][18], and what Microsoft has to stand up inside that window is its own compliance side: annual certification, audit rights running against it, a fixed deadline to remediate security issues ^[9], and breach notice inside 72 hours ^[6]. Smith described 30 pages reducing to three words, privacy, safety and transparency ^[8], which works out to three pages per enforceable protection ^[21].

Read the ten as a bill of materials for products Microsoft has decided not to sell into American classrooms: student data cannot enter model training, students cannot be tracked, no AI decision about a student stands without a human reviewing it ^[4], nothing goes to advertisers, and companion chatbots for students are barred outright ^[6]. Those were among the reasons more than 250 child-safety experts and organisations asked for a school AI moratorium in April ^[16].

The pressure on rivals is coming from the unions rather than from Microsoft. Weingarten said both OpenAI and Anthropic have expressed willingness to do this kind of agreement and that she hopes they sign soon ^[10]; neither responded to Fortune's requests for comment ^[11]. Her account of why the union went through contracts at all is that HIPAA and FERPA never envisioned AI ^[12], and COPPA, which requires parental consent before a company collects data from a child under 13, predates generative AI by more than two decades and says nothing about how models are trained ^[13].

Mulgrew, whose New York City local has roughly 200,000 members ^[14], said in the same appearance that his union does not believe K-8 AI use is appropriate at this point and that teachers should work out what fits each grade level ^[15]. That is the division of labour the standard implies: the contract governs the data, the teacher governs the grade, and the nation's largest school system ^[19] is out of the market for a year regardless ^[1].

One path: a rival signs, and ten protections drafted by a competitor become the entry price for district business. Another: nobody signs, leaving Microsoft holding a clause it can quote in a bid. Or districts simply take the free default, never paste the clause in, and the standard ends up as an audit calendar with no plaintiff. The evidence points at the second, because the only parties on the record asking OpenAI and Anthropic to join are the two unions ^[10]. A signed rival version before November 1 would end that read; so would a district publishing an executed clause with a damages figure in it.