Student Data Privacy and AI: What Microsoft Now Promises

How FERPA, COPPA and Microsoft’s new AI Safety and Privacy Standard protect student data, plus red flags and a 10-clause test for any AI vendor.

AI privacy in schools: students gathered around a laptop in a classroom

Last updated: November 7, 2026

Reviewed by a privacy professional (CIPP/US) on November 10, 2026.

US student data privacy law was built for filing cabinets, not chatbots. FERPA goes back to 1974, and as AFT President Randi Weingarten put it, “HIPAA and FERPA never envisioned the advent of AI.” Since November 1, 2026, schools have had a new way to close that gap: Microsoft contract terms that ban training AI on student data and keep a person in charge of high-risk decisions.

Short answer: FERPA only lets schools share student records with AI vendors under conditions like the school official exception, and it says nothing specific about AI training. Microsoft’s AI Safety and Privacy Standard, open to US districts from November 1, 2026, adds terms a district can enforce: no training on student data, no sales or ads, human review and breach notice within 72 hours.

Key takeaways

  • Under FERPA, an AI vendor acting as a school official can use student data only for the purposes it was shared for.
  • Microsoft’s standard lists 10 protections, and districts can add it to existing agreements without renegotiating or waiting for renewal.
  • For services aimed at children under 13, most provisions of the amended COPPA Rule took full effect on April 22, 2026.
  • When K-12 Dive reported in September, Microsoft was the only company that had signed the unions’ standard.

What student data privacy means once AI is in the classroom

Student data privacy means protecting personal information about students, from grades and attendance to anything else that identifies them, against collection, use or disclosure that they or their parents haven’t agreed to. AI tools push that definition in three directions.

  • More data. A chatbot conversation, an essay draft or a voice recording can all hold personal details a gradebook never did.
  • New uses. A vendor might want to use that data to train or improve its models, a use most privacy rules never anticipated.
  • Automated decisions. AI can flag a student as at-risk, score an essay or recommend a placement, which raises the question of who checks the output.

That is why the student data privacy debate has shifted from whether schools use AI to what the contract says. Federal law sets the floor. Contracts are where districts can raise it.

FERPA and AI: what the law actually covers

The Family Educational Rights and Privacy Act (FERPA) lets parents see their children’s education records, ask for corrections and have some say over disclosure of personally identifiable information in them, according to the Department of Education’s Student Privacy Policy Office. Those rights move to the student at 18 or when they start at a postsecondary institution.

FERPA and AI cross paths at one provision, the school official exception. It allows a school to share records with an outside party such as an AI vendor without parental consent, but only when the vendor meets four conditions:

  1. It performs an institutional service or function the school would otherwise use employees for.
  2. It is under the school’s direct control with respect to the use and maintenance of education records.
  3. It uses personally identifiable information only for the purposes for which it was disclosed, and is limited in redisclosing it.
  4. It meets the criteria in the school’s annual FERPA notice for school officials with a legitimate educational interest.

What FERPA doesn’t do is mention AI. Model training, automated decisions and chatbot logs aren’t in it, and it has no breach-notice deadline of its own. So districts have to turn condition 3, use only for the authorized purpose, into concrete contract language. That’s the student data privacy gap the Microsoft standard is trying to plug.

COPPA and other rules that apply

The Children’s Online Privacy Protection Act applies to online services aimed at children under 13. The FTC’s amended COPPA Rule took effect June 23, 2025, and most operators had until April 22, 2026, to comply. Interestingly, the FTC decided not to finalize its proposed rules on ed tech and the role of schools, pointing to possible clashes with pending FERPA changes. Plenty of states have their own student data privacy laws too, so check yours before you sign an AI contract.

Here’s how the main student data privacy rules stack up once AI tools are in the picture:

RuleWho it applies toWhat it requiresWhy it matters for AI
FERPASchools and their school officials, including qualifying vendorsConsent before disclosing personal information from education records, with exceptionsAI vendors must meet the school official conditions and use data only for authorized purposes
COPPA (amended 2025)Operators of online services directed to children under 13Verifiable parental consent and data protectionsAI tools for younger students need a consent and retention plan
Microsoft AI Safety and Privacy StandardUS districts that add it to Microsoft agreements10 contractual protectionsBans training on student data and requires human review
Laptop on a classroom desk, where student data privacy rules apply to AI tools

Microsoft’s AI Safety and Privacy Standard: the 10 protections

On September 9, 2026, the American Federation of Teachers, the United Federation of Teachers and Microsoft announced the standard. Microsoft’s fact sheet sets out ten student data privacy and safety protections:

  1. No training on student or educator data. Student, educator and customer data can’t be used to improve AI models, except under a narrow safety and security carve-out.
  2. Data minimization by design. Products take only the data they need, and riskier features like tracking and biometrics are limited and reviewed.
  3. Customer data ownership. Data can’t be sold or used for ads or product development, and the customer controls export, retention and deletion.
  4. Human oversight. As the fact sheet puts it, “AI supports, people decide.” High-risk decisions get human review, and AI companions are banned.
  5. Real accountability. Providers hold certifications and report breaches within 72 hours. Violations get fixed, or AI use is paused or stopped.
  6. Security that matches the stakes. That means encryption, access controls and independent testing, plus incident-response plans that cover subprocessors.
  7. Consent and transparency for families. Families get plain-language guides on how products work and how privacy, safety and security are handled.
  8. Equity and accessibility. Providers have to disclose and close gaps, offer equal features and pay for an independent study.
  9. No feature creep or lock-in. Customers decide on significant new features and data uses, and they can export data and change providers freely.
  10. Long-term responsibility. Data is kept only as long as it’s needed, the training and sale bans never expire, and deletion and breach duties outlast the contract.

How districts adopt it

Microsoft says the protections are available to US school districts beginning November 1, 2026, with no contract renegotiation or renewal required, and are contractually enforceable. UFT President Michael Mulgrew said the agreement lets districts end agreements and seek damages from companies that break the rules. Microsoft also said a similar framework had been adopted by the Los Angeles Unified School District, New York State United Teachers and New York City Public Schools.

As of K-12 Dive’s report on the announcement, Microsoft was the only company to sign the unions’ standard. Weingarten said she hoped OpenAI and Anthropic would follow. We found no announcement of other signatories as of November 7, 2026. In our view, that makes it one of the most concrete student data privacy commitments from a major AI vendor so far.

Smiling teacher sitting at a desk in a classroom with a chalkboard behind her

FERPA and AI terms vs. Microsoft’s standard

Here is the student data privacy gap side by side. This comparison is our own reading of the two documents, not legal advice.

IssueFERPA on its ownWith Microsoft’s standard
Training AI on student dataNot named; vendor use limited to the purpose of disclosureBanned, apart from a narrow safety and security exception
Selling data or using it for adsNot named; covered only through use limitsBanned permanently
Breach noticeNo FERPA deadlineWithin 72 hours
Human review of AI decisionsNot addressedRequired for high-risk decisions
AI companionsNot addressedProhibited
New features using student dataNot addressedCustomer controls material new features and data uses
Leaving the vendorNot addressedExport data and switch providers freely
EnforcementDepartment of EducationContract remedies, including pausing or ending AI use

Put side by side, the split is obvious. FERPA decides who can get student records. The standard decides what a vendor can do with them afterward, and for student data privacy in the AI era that second question is the bigger one.

Student data privacy red flags in AI tools

Most student data privacy trouble with AI begins in the terms of service, not with a breach. Look for these red flags before anyone in your district clicks accept:

  • Training language. Terms that let the vendor use your content or inputs to improve its services, which usually means model training.
  • Consumer accounts. Teachers signing up for a free tier with a personal email, which puts student data under consumer terms instead of a school contract.
  • No deletion timeline. A privacy policy that promises deletion when data is no longer needed, with no number of days attached.
  • Vague subprocessors. No list of the other companies that receive the data.
  • One-sided changes. A clause that lets the vendor change its terms by updating a web page.

Each one is a student data privacy risk the 10-clause test below is designed to catch. FERPA and AI vendor terms collide at exactly these points: the school official exception only works if the school keeps direct control, and a teacher’s consumer account gives it none.

The 10-clause AI privacy test for any vendor

You don’t need to be a school, or a Microsoft customer, to use these terms. Here is a checklist we built from the standard. These are the student data privacy terms worth demanding. Put the questions to any AI vendor, in writing:

  1. Will you use our data to train or improve models, and will you put the answer in the contract?
  2. Will our data ever be sold, used for ads or used for product development?
  3. What data do you collect that the service doesn’t strictly need?
  4. Who controls export, retention and deletion?
  5. Where does a person review high-risk outputs?
  6. What is your breach-notice deadline? Is 72 hours possible?
  7. Which subprocessors handle our data, and do the same terms bind them?
  8. Will you tell us before turning on new features that use our data?
  9. Can we export everything and leave without penalty?
  10. Do deletion and breach duties survive the end of the contract?

Our AI vendor questionnaire turns points like these into a form you can send as-is, and our SOC 2 guide shows how to check the security evidence a vendor hands over.

A 5-step AI policy for schools

A school AI policy is where student data privacy turns into day-to-day practice. It can be short, as long as it answers five questions:

  1. What AI tools are in use? List all of them, including free tools teachers picked up on their own.
  2. What data goes in? Classify it: directory information, education records, sensitive data. Ban sensitive categories from general-purpose tools.
  3. Does each vendor meet the FERPA school official conditions? Check the contract for direct control, use limits and redisclosure limits.
  4. Which contract terms do you require? Treat the 10-clause test above as the floor, and add the Microsoft standard to any Microsoft agreement.
  5. Who tells families? Post a plain-language list of the tools you use and what each one does with student data.
Students working together on laptops, where student data privacy rules apply

Our Take / Expert Perspective

The important word in Microsoft’s announcement is contractually. Student data privacy promises on a website can be rewritten whenever a product team needs new data. A clause in a signed agreement cannot, at least not without the customer’s say. That is what the teachers’ unions secured, and it is the bar other buyers will now point to.

For student data privacy, the real question isn’t whether FERPA is out of date. It’s who covers the gap until the law catches up, and right now the answer is procurement. One big vendor has put a training ban, 72-hour breach notice and a free exit in writing, which gives every district bargaining with another AI company a precedent to point to. Don’t be surprised to see it show up in business contracts as well, the way disclosure rules are spreading for pricing that uses personal data.

The losers are vendors that rely on vague terms of service. Our forecast, and it is a forecast: within a year, districts will treat a written no-training clause as a baseline requirement, not a perk.

One caution. The standard covers what Microsoft promises to districts that add it. It changes nothing in your own Microsoft contract unless your agreement includes it.

Our recommendation: if you run a district, add the standard to your Microsoft agreements and send the 10-clause test to every other AI vendor before the next school year.

Frequently Asked Questions

What is student data privacy?

Student data privacy is the protection of personal information about students, such as grades, attendance, health and behavior records, from collection, use or disclosure without proper consent. In the US it is governed mainly by FERPA, COPPA for children under 13, state laws and the contracts schools sign with vendors.

Does FERPA apply to AI tools?

Yes, when a school shares personally identifiable information from education records with an AI vendor. Most of the time that happens under the school official exception, which generally means the vendor has to qualify, stay under the school’s direct control and use the data only for the reason it was shared. FERPA doesn’t mention AI training.

Can AI companies train models on student data?

FERPA limits a vendor’s use of student records to the authorized purpose but doesn’t name model training. Microsoft’s AI Safety and Privacy Standard explicitly bans training on student, educator and customer data, apart from a narrow safety and security exception, for districts that adopt it.

Is Microsoft’s AI Safety and Privacy Standard a law?

No. It’s a set of contract commitments districts can add to their Microsoft agreements from November 1, 2026, enforceable through that agreement rather than through any statute.

Have other AI companies signed the standard?

As of K-12 Dive’s September 2026 report, Microsoft was the only signatory, and the AFT president said she wanted OpenAI and Anthropic to join. We found no other signatories announced as of November 7, 2026.

Does this standard apply to businesses?

No. Microsoft offers it to US school districts. Businesses can still ask their own AI vendors for similar terms, using the 10-clause test in this article.

This article is for general information only and is not legal advice. Laws differ by state, province and country; consult a qualified attorney for your situation.

Sources

Featured photo: Vitaly Gariev on Unsplash.

Other photos: Thomas Park, Vitaly Gariev and Brooke Cagle on Unsplash.

Sachin Garg Avatar

Leave a Reply

Your email address will not be published. Required fields are marked *

About aiandblog

Practical reporting on AI, SaaS, compliance and finance. We cite primary sources and correct mistakes in the open. Read our editorial policy.