FERPA and AI in Education: A Guide for Administrators

The legal mechanism that lets a district hand student data to an AI vendor at all is FERPA's "school official" exception, codified at 34 CFR § 99.31(a)(1), and most board policies were written before generative AI made that exception load-bearing in a new way. An administrator's real job isn't memorizing FERPA's text; it's making sure the district's contracts, board policy, and staff training actually satisfy the exception's three specific conditions.

Quick Answer: FERPA lets a district disclose education records to an AI vendor without individual parental consent only through the "school official" exception (34 CFR § 99.31(a)(1)(i)(B)), which requires the vendor to perform a service the school would otherwise do itself, remain under the district's direct control, and be contractually bound by the same use-and-redisclosure limits as an employee. Administrators are responsible for three things this article covers in depth: verifying that test actually holds for each AI vendor, updating board policy and the annual FERPA notification to reflect it, and training staff so the policy is followed rather than quietly bypassed.

This is a deliberately narrower, governance-focused companion to the broader legal landscape in our AI, FERPA & COPPA: The Complete School Data-Privacy Guide (2026). Where that guide covers what FERPA and COPPA require in general, this one is written specifically for the administrator deciding how a district's policy, contracts, and training obligations should actually work.

Why FERPA Suddenly Needs an Administrator-Specific Answer

Generative AI tools ingest identifiable student work in ways FERPA's 1974 drafters never anticipated, which makes the outsourcing question sharper than it was for earlier edtech. A gradebook stores scores; an AI writing-feedback tool reads, processes, and sometimes retains a student's full essay to generate a response.

District policy has been racing to catch up. EdWeek Market Brief's 2026 survey of school-system technology leaders found roughly 80% of districts now have written AI guidelines in place, up from about 57% the year before, while the share with no guidance at all fell from 43% to 21% over the same period (EdWeek Market Brief, 2026). That's real progress, but "having guidelines" and "having board-adopted policy that correctly applies FERPA's school-official test" are not the same milestone.

Part of the gap is structural. FERPA itself has no private right of action, a point the U.S. Supreme Court settled in Gonzaga University v. Doe, 536 U.S. 273 (2002), which means a family generally cannot sue a district directly over a FERPA violation. Enforcement runs through the U.S. Department of Education, which can, after giving a district the chance to fix a problem, terminate its eligibility for federal funding, a penalty the Department has authority to impose but has never actually used.

That combination, a serious-sounding penalty rarely enforced, is exactly why administrators sometimes treat FERPA-AI questions as lower urgency than they should be. The reputational and state-law exposure sitting underneath the federal penalty is where the real risk lives.

The School Official Exception: The Legal Mechanism That Makes AI Vendors Possible

FERPA generally requires parental consent before disclosing a student's education record to a third party, but the school-official exception lets a district skip that consent when an outside vendor is treated as if it were an employee. This exception is the entire legal basis for using any AI vendor with real student data at all.

The Three-Part Test Under 34 CFR § 99.31(a)(1)(i)(B)

A contractor or vendor can be treated as a school official, and therefore covered by the exception, only if all three conditions hold:

1. The vendor performs an institutional service or function the district would otherwise perform itself, such as content generation, grading support, or tutoring assistance a teacher would otherwise do unaided.
2. The vendor remains under the district's direct control with respect to how it uses and maintains education records, meaning the district sets the terms, not the vendor.
3. The vendor is bound by the same use-and-redisclosure limits FERPA imposes on employees, specifically the restrictions in 34 CFR § 99.33(a), which bar redisclosing personally identifiable information without authorization.

A vendor's public-facing terms of service almost never satisfies condition two or three by itself. Only a negotiated Data Processing Agreement, reviewed by the district and specific to that vendor, actually establishes "direct control" in a way that would hold up if ever questioned.

What "Legitimate Educational Interest" Means for an AI Tool Specifically

A school official, human or vendor, may only access records where there's a legitimate educational interest in that specific information, not blanket access to every student record in the district. For an AI vendor, that means scoping data access to what the specific tool actually needs.

A districtwide AI writing tool used only in English classes doesn't need access to a student's disciplinary file or health records. Practically, this means procurement should ask each vendor to specify exactly which data fields the tool ingests, not accept a broad "student information" description that would fail this test if scrutinized.

The Annual Notification Requirement Administrators Often Miss

FERPA requires a district to specify, in its annual notification of FERPA rights, the criteria it uses to determine who qualifies as a school official and what counts as a legitimate educational interest, per 34 CFR § 99.7(a)(3)(iii). Many districts' annual notices were last substantively updated before AI vendors were part of the conversation.

This isn't a minor paperwork detail. If a district's annual notice still describes "school officials" only in terms of employees and never mentions contracted vendors, a parent or auditor has a reasonable argument that the district hasn't actually extended the exception to cover AI tools in the way it's using them in practice.

Two things worth checking during any policy refresh:

• Does the annual notice explicitly describe the criteria for treating a vendor as a school official, not just employees?
• Is that language consistent with the actual contracts on file, so the notice isn't describing a stricter or looser standard than what's actually being applied?

Board Policy Considerations: What Belongs in Written District Policy

A board policy that only says "AI tools may be used at teacher discretion" leaves the school-official test unaddressed entirely. A policy built to actually satisfy FERPA's structure should cover several specific elements.

1. A defined AI vendor approval process, naming who reviews a new tool (IT, legal, a designated student-data-privacy officer) before it reaches a classroom.
2. A requirement that a signed DPA exist before any AI tool processes identifiable student data, not just a consumer terms-of-service acceptance.
3. Explicit criteria for "school official" designation, matching the three-part test above, so the annual notification and the actual contracts stay aligned.
4. Data-minimization defaults, limiting what student information any given AI tool can access to what its specific function requires.
5. A prohibition on entering personally identifiable student information into unapproved consumer AI tools, addressing the shadow-AI risk that shows up long before any formal policy catches it.
6. An annual re-verification requirement, since a vendor's data practices, subprocessors, and model provider can all change after the initial approval.

Each of those elements maps back to a specific gap the school-official test creates if left unaddressed.

The National Education Association publishes a sample school board policy on AI issues that districts can use as a starting reference point, and CoSN's K-12 AI guidelines offer a broader governance framework worth reviewing alongside legal counsel (NEA; CoSN, 2025). Neither substitutes for district-specific legal review, but both are useful starting drafts rather than starting from a blank page.

Staff Training Obligations: Turning Policy Into Practice

A board policy that staff have never been trained on functions the same as no policy at all. FERPA's protections depend on the people actually handling student data, teachers, aides, and administrative staff, understanding what counts as a disclosure and when.

Training should cover, at minimum:

• What counts as an education record when entered into an AI tool. A student's essay, grades, IEP details, or behavioral notes typed into a chat interface can all qualify, even informally.
• The difference between an approved, contracted AI tool and a personal account. Shadow AI use, a well-meaning teacher's personal ChatGPT account, is the most common way student data ends up outside any DPA's protection.
• How to verify a tool has cleared the district's vendor-approval process before using it with identifiable student data, not after.
• What to do if student data appears to have been mishandled, including who to report it to and how quickly.

CoSN, working with AESA, SETDA, and AASA on the Building Capacity for Generative AI in K-12 Education initiative, has built structured professional-learning resources specifically for this gap, supported by a Bill & Melinda Gates Foundation grant (CoSN, 2025). Districts without in-house capacity to build this training from scratch have a real, external option rather than needing to originate it internally.

Training also needs a cadence, not just a one-time rollout session. New staff joining mid-year, a new AI tool added to the approved list, or a policy revision after a board meeting are all moments that should trigger a short refresher, not wait for the next scheduled annual training. A district that trains once at policy adoption and never again tends to see compliance quietly erode as staff turn over and memory of the original session fades.

A Practical Scenario: Applying the School-Official Test to a New AI Tool

Say you're an assistant superintendent for curriculum reviewing a new AI tool that drafts differentiated reading passages for a Grade 5 social studies unit, and a principal has already piloted it informally with one class. Before it moves to a districtwide recommendation, run it through the three-part test directly.

• Institutional service: Confirm the tool is doing work a teacher would otherwise do manually, drafting differentiated text at multiple reading levels, rather than something categorically new the district wouldn't otherwise provide.
• Direct control: Confirm a DPA exists specifying exactly what student data (if any) the tool accesses, how long it's retained, and that the district, not the vendor, can require deletion.
• Redisclosure limits: Confirm the contract extends those same restrictions to any subprocessor, including whichever foundation model powers the tool underneath.

Tools built around a teacher-managed structure can make this specific test easier to satisfy, since less individual student data changes hands in the first place. EduGenius, for example, is designed around a teacher configuring a class profile (grade level, subject, ability range) and generating differentiated materials from it, rather than students interacting with the model directly, which narrows what a school-official analysis actually has to account for. That's a design characteristic to weigh during review, not a substitute for confirming the DPA and contract terms actually hold up.

For the COPPA-specific consent and retention questions that apply once younger students are involved, our COPPA compliance checklist for AI tools in schools covers that ground in detail, and for a security-focused companion to this legal review, see How Schools Can Use AI Without Violating SOC 2. For how these same governance questions apply to a specific classroom-facing product, see Is Nearpod COPPA Compliant?

Related Reading

• AI, FERPA & COPPA: The Complete School Data-Privacy Guide (2026)
• Diffit and Student Data Privacy: What Teachers Should Know
• A COPPA Compliance Checklist for AI Tools in Schools
• Is Nearpod COPPA Compliant?
• How Schools Can Use AI Without Violating SOC 2
• SchoolAI vs Khanmigo: Which Is Better for Teachers?