Can schools put student data into generative AI systems?

It depends on the product, contract, applicable law, school policy, and how the data is handled. Schools should not assume that a consumer AI account offers education-grade protections.

What does "not used for training" mean?

It generally means information submitted by the institution is not used to improve the provider's underlying AI models. Schools should verify the exact contractual language and exceptions.

Is anonymized data risk-free?

No. De-identification reduces certain privacy risks but does not automatically eliminate all risks. Schools should understand how de-identification is performed and how aggregated information may be used.

AI Student Data Privacy: 2026 Vendor Checklist

What Should Schools Require From an AI Vendor?

At minimum, a school evaluating an AI vendor should know:

what data is collected,
why it is collected,
whether it is used to train AI models,
who can access it,
how long it is retained,
how it can be deleted,
whether users are tracked,
who owns uploaded and generated information,
which subprocessors receive data,
and whether consequential decisions require human review.

These should be contract questions, not assumptions.

On September 9, 2026, Microsoft, the American Federation of Teachers, and the United Federation of Teachers announced a National AI Safety & Privacy Standard for schools. Among its requirements are restrictions on using student and educator data for model training, data monetization, tracking, and AI decisions made without human oversight.

The significance is larger than one vendor.

Expectations are changing.

1. Is Student Data Used to Train AI?

This should be one of the first questions.

Schools should distinguish between:

operating the service,
improving the service,
safety monitoring,
analytics,
and model training.

Those are not necessarily the same data uses.

A vendor should provide a clear answer in plain language.

2. What Data Does the Vendor Actually Store?

"FERPA compliant" is not a complete data architecture.

Schools need to understand the actual data flow.

Does the system store:

student names?
email addresses?
grades?
assessment responses?
parent information?
disability-related information?
behavioral information?
chat histories?
uploaded documents?

A product that minimizes personally identifiable information may present a materially different risk profile from one that centralizes extensive student data.

3. What Is the Retention Policy?

Data should not exist indefinitely simply because deleting it is inconvenient.

Ask:

How long is production data retained?
What happens after a contract ends?
Is there an export period?
When is deletion completed?
How long may encrypted backups retain deleted information?

Retention should be explicit.

4. Who Owns the Data?

The school should understand ownership of:

student data,
uploaded curriculum,
assessments,
teacher-created content,
generated reports,
aggregated analytics.

The contract should make these distinctions clear.

5. Which Subprocessors Have Access?

Modern AI products often depend on infrastructure and model providers.

A school should know which third parties process its information and what protections apply downstream.

Vendor evaluation should therefore extend beyond the visible product interface.

6. Is Human Oversight Required?

Microsoft's September 2026 education principles state that consequential decisions should remain subject to meaningful human oversight.

In education, this matters particularly for:

placement,
disciplinary actions,
interventions,
academic progression,
special education decisions,
high-stakes assessment interpretations.

AI may identify signals.

Educators should retain decision authority.

7. Can the School Leave?

Good governance includes an exit plan.

Ask:

Can we export our information?
In what format?
What happens to our data afterward?
How quickly is access terminated?
Can the vendor retain aggregated information?
What happens if the company is acquired?

Privacy is not only about entering a vendor relationship.

It is also about leaving one.

The Omnix360 Vendor Data Governance Checklist

Before approving an AI system, require clear answers across seven areas:

Collection → Purpose → Training → Access → Retention → Ownership → Deletion

If any one of these is vague, the school's risk is also vague.