Microsoft bans AI companions under a new school privacy and safety standard negotiated with the American Federation of Teachers and New York City’s United Federation of Teachers. The agreement also restricts training on student and educator data and requires human review for high-risk decisions.

 

The school AI standard establishes five concrete protections:

  • No model training on student or educator data.
  • Limits on the data Microsoft collects.
  • Plain-language disclosures for families.
  • No AI companion applications in schools.
  • Human review for high-risk decisions.

 

Background Reading

 

Microsoft Bans AI Companions and Model Training on School Data

The agreement addresses two questions that have become central to classroom AI: what information a provider may use, and what kinds of automated relationships it may encourage. Microsoft committed not to use student or educator data to train AI models, according to reporting by The Verge on the published standard.

 

That restriction separates operating a contracted service from improving a general model with information generated inside schools. It does not mean the systems process no data. Microsoft also committed to minimize collection and explain its tools to students, families and educators in plain language.

 

The ban on AI companions draws a boundary around products designed to form ongoing social or emotional relationships with users. Schools may still use permitted AI tools for instructional or administrative work, but the standard rejects companion-style applications within Microsoft’s covered education services.

 

Human review is required for high-risk decisions. That principle matters when an automated system influences outcomes such as access, discipline, placement or other consequential judgments. The agreement does not turn a model’s recommendation into an authorized final decision merely because the software can generate one quickly.

 

November 1 Extends the Standard Across US School Districts

Microsoft President Brad Smith said the 30-page agreement is already binding between Microsoft, the AFT and the UFT. He also said it will take effect on November 1 for every U.S. school district that does business with Microsoft, without districts needing to opt in separately.

 

That timetable gives the announcement a wider reach than a pilot involving a single city. Microsoft sells cloud, productivity and AI services throughout American education, so applying common terms across those relationships could move privacy and safety commitments into procurement without waiting for each district to negotiate the same language from scratch.

 

The mechanism remains contractual rather than legislative. It governs Microsoft’s covered relationships and can be enforced through the agreement’s provisions, including audit rights described by Smith. It does not automatically bind competing AI providers, change state law or determine whether a district permits a particular classroom use.

 

District leaders will still need to map the standard onto local systems and practices. A contractual promise is most useful when administrators know which products it covers, where data flows, how long information is retained, who can review logs and what happens when an audit identifies a breach.

 

AFT and UFT Put Educators Into AI Contract Negotiations

The American Federation of Teachers is the second-largest teachers union in the United States, while the UFT represents educators in New York City. Their participation places workforce and student concerns inside the vendor agreement instead of leaving the terms solely to a technology company and individual procurement offices.

 

AFT President Randi Weingarten has argued that enforceability is essential because broad statements about responsible AI do not create the same obligations as contract language. Microsoft’s standard is framed around privacy, safety and transparency, with the ten principles intended to translate those themes into operational commitments.

 

The agreement arrives as major school systems are reassessing student-facing AI. New York City and Los Angeles recently imposed one-year restrictions on many such tools, creating time to examine safeguards and acceptable uses. Microsoft’s approach offers contract terms, but each district retains responsibility for deciding when technology belongs in teaching.

 

Educators also need practical ways to challenge an automated output. Human review has little value if staff cannot see the evidence behind a recommendation, understand which data shaped it or override it without penalty. Implementation will determine whether the principle functions as a real safeguard or a procedural label.

 

Audits Will Test Microsoft’s School AI Privacy Standard

The November rollout creates several measurable tests. Districts can examine whether Microsoft’s documentation clearly identifies covered data, whether companion features remain excluded, whether high-risk workflows require meaningful human action and whether audit provisions provide enough evidence to investigate complaints.

 

Families will also need accessible explanations of which tools are active and what choices are available. A plain-language notice should describe purpose, inputs, retention and decision authority rather than merely restating that a product uses artificial intelligence.

 

The standard could influence future education contracts if districts ask other vendors to match its protections. That effect is not automatic, and competitors may define covered services or high-risk decisions differently. Comparable terms will require procurement teams to examine definitions, exceptions and enforcement rather than compare only headline promises.

 

Microsoft’s agreement gives schools a specific baseline before the next wave of classroom AI deployments. Its significance will depend on how consistently the company applies the restrictions, how easily districts can audit compliance and whether educators and families can use the enforcement mechanisms when problems occur.