The short answer. When a university lets an AI tool work with education records, FERPA treats the vendor as an outside party that can receive student data without consent only if it qualifies as a “school official.” That means four conditions, including that the institution keeps direct control over how the records are used and maintained (34 CFR 99.31(a)(1)(i)(B)). Most consumer AI tools can’t meet that bar. A campus AI program is defensible when it runs on tools the institution controls, limits each user’s AI to the records they have a legitimate educational interest in, and shows which source produced every answer.
Students moved first
In a Lumina Foundation-Gallup survey of 3,801 enrolled US college students conducted in October 2025, 57% said they use AI in their coursework at least weekly, and about one in five use it daily. More than half said their school discourages AI use (42%) or prohibits it outright (11%) (Gallup, 2026). Faculty and staff aren’t far behind, and they’re the ones with access to grades, advising notes and accommodation files.
That’s where the FERPA exposure sits. A student pasting their own essay into a chatbot is an academic-integrity question. An advisor pasting a caseload of student records into the same chatbot is a disclosure.
The four conditions, and where campus AI breaks them
The Department of Education’s Student Privacy Policy Office lists four conditions for an outside party to count as a school official (Student Privacy Policy Office, FAQ):
| FERPA condition | Where campus AI tools fall short |
|---|---|
| Performs an institutional service or function the institution would otherwise use employees for | Usually met: drafting, summarizing and answering questions are staff work |
| Is under the institution's direct control with respect to the use and maintenance of education records | Consumer tools set their own retention and data-use terms; the institution can't direct them |
| Uses personally identifiable information only for the purposes of the disclosure (34 CFR 99.33(a)) | Terms that allow use of inputs for other purposes conflict with this |
| Meets the institution's own criteria for a school official with a legitimate educational interest | A tool that lets any user query every record can't enforce this per person |
The fourth condition gets the least attention and causes the most trouble. Even a properly contracted AI tool fails it if a financial aid counselor’s question can reach disciplinary files. FERPA’s legitimate-educational-interest test is per person, so the AI has to enforce it per person.
An AI evidence review workflow that holds up
A written policy tells people what they may do. An evidence review workflow shows, after the fact, that they did it. Five steps:
- Load the authoritative documents into governed spaces. Catalog policies, program requirements, advising guides and approved templates, and decide who can reach each set.
- Match AI access to legitimate educational interest. An advising space and a financial aid space, for example, each with its own membership.
- Require a source on every consequential answer. If the AI can’t cite an institutional document, the answer doesn’t go into a student-facing decision.
- Review the evidence, not just the answer. The staff member opens the cited source before acting on it.
- Keep a record of which documents were used and when. When a student appeals or an auditor asks, you produce the record instead of reconstructing it.
The test for a campus AI tool is the same as for a campus employee: can you show what they looked at, and were they allowed to look at it?
Where Mithra fits
Mithra® is a governed AI workspace for an institution’s own documents. Files go into permissioned Library spaces, and space membership decides whose questions can reach which documents. That’s how the legitimate-educational-interest line gets enforced where the AI reads, not just in the handbook. Each file is tokenized on upload with a Token ID, a timestamp, its owner and its metadata.
Every answer carries a Trusted, Partial or Untrusted label, or General Knowledge when no institutional document applies, and View Evidence opens each cited source. With “Use only your data sources” turned on, questions the institution’s documents can’t answer are declined rather than filled in from general knowledge. Admins can pull outdated catalogs or retired policies out of scope by metadata value, and each document’s Data Lineage view logs when it was cited or viewed as evidence. AI Threat Protection runs on every query and treats documents as data, never as instructions, so text hidden inside an uploaded file can’t redirect the AI.
Mithra doesn’t connect to your student information system or LMS, and it doesn’t decide which document is correct. The institution decides what goes into each space. Mithra keeps every answer tied to it.
Frequently asked questions
Can faculty use ChatGPT with student records under FERPA?
Only if the tool qualifies as a school official under 34 CFR 99.31(a)(1)(i)(B), which requires the institution to keep direct control over how the records are used and maintained. Consumer AI terms rarely allow that, so most institutions prohibit entering education records into ungoverned AI tools.
What makes an AI tool FERPA compliant?
No tool is FERPA compliant on its own; institutions comply. A tool supports compliance when the institution controls how it uses records and when it enforces access by legitimate educational interest, person by person.
How do universities audit AI use?
By keeping a record of which institutional sources each AI answer relied on and who could reach them. That record has to be captured when the answer is generated.
About Mithra Technologies Inc.
Mithra Technologies Inc. is building the trust infrastructure for enterprise AI. Centered on its cryptographic Single Source of Truth (SSOT®), Mithra enables organizations to establish verifiable AI through trusted enterprise content, cryptographic proof, complete data lineage, provenance and governance. Designed to work across any AI model, Mithra helps enterprises deploy AI that is more transparent, explainable, auditable and resilient while reducing the risks associated with misinformation, unauthorized data usage, cyber threats and the emerging quantum era.