Schools are adopting AI tools faster than they are writing rules for them, and that gap is where the announcement of 9 September lands. Microsoft, the American Federation of Teachers and the United Federation of Teachers published an AI safety and privacy standard that US school districts can incorporate into their Microsoft agreements. It is a contract, not a law, and understanding that distinction is the key to judging what it actually does.
A contractual route to safeguards
Because the standard is contractual, it applies where a district chooses to adopt it and signs on. It does not bind every education technology company, and it does not override existing law. That is both its strength and its limit: a district can move quickly without waiting for legislation, and a district that does not adopt it changes nothing. The mechanism is voluntary and opt-in.
What the provisions cover
The announcement addresses the handling of school data, including restrictions on selling it or using it to train AI models, along with transparency and human oversight. For schools, these are the questions that matter most. Student data is sensitive, and the prospect of it feeding a commercial model is exactly the concern that has made educators cautious about AI adoption.
Why data training is the sensitive issue
Using school records to train AI models raises obvious privacy questions, because the data concerns children and is collected for educational purposes rather than for product development. Even when data is aggregated or anonymised, the risk of re-identification and the question of consent remain. A contractual commitment not to use school data in this way addresses the most acute concern, and it is a commitment made by the company about its own practices.
Human oversight is easy to promise and harder to define. Does it mean a teacher reviews every AI-generated recommendation? A district-level committee approves deployments? A mechanism for appeal when an AI system makes a consequential decision about a student? The value of the standard depends on how specifically it answers these questions, which is why the full agreement is more informative than the announcement.
The national framing problem
The initiative is described in national terms, which can suggest it applies everywhere. It does not. Education in the United States is largely a state and district matter, and a private agreement between a company and unions cannot create a nationwide rule. Schools in districts that do not adopt the standard are unaffected. This is a clarification that prevents misunderstanding.
What schools should ask
A district considering adoption should ask which products and services the standard covers, how compliance is verified, what happens if the company changes its practices, and how disputes are resolved. Contracts are only as good as their enforcement mechanisms, and a commitment without an audit or remedy is a statement of intent rather than a safeguard.
AI in education is expanding quickly, driven by promises of personalised learning and administrative efficiency. The risks – bias in assessment, surveillance of students, erosion of teacher judgement – are documented in research and in the concerns of educators. A voluntary standard is one response among several, alongside regulation, procurement rules and professional guidance. It is a useful piece of the puzzle, and not a complete answer.
How the unions got involved
Teachers’ unions have been among the more active voices on classroom technology, and their involvement here is significant. Unions bring a different perspective from technology vendors: they represent the people who will use the tools and who will bear the consequences when a system is wrong. Involving them in drafting a standard gives the resulting commitments more credibility with educators, and it also means the standard reflects concerns about workload, professional judgement and student privacy that a company might not prioritise on its own.
What other places are doing
Several countries and states have taken a more regulatory route, issuing guidance, restricting certain tools or requiring impact assessments before AI systems are deployed in schools. The approaches differ, and they share a common difficulty: technology changes faster than rules. A contractual standard can be updated more quickly than legislation, which is one argument in its favour, and it also lacks the force of law, which is the counterargument.
Voluntary standards work when adoption is widespread and when non-adoption carries a cost. If only some districts sign on, the effect is uneven, and companies may offer weaker terms where they face less pressure. The strongest outcome would be a standard that becomes a de facto requirement, because districts expect it and vendors compete on compliance. Whether this initiative reaches that point will depend on take-up over the next few years.
This report draws on Microsoft.
Image: Billjones94 · CC BY-SA 4.0 · via Wikimedia Commons.