FERPA and AI: Compliance Checklist for Universities
FERPA compliance for AI systems in higher education requires treating every AI agent query against a student information system as a disclosure event governed by 34 CFR Part 99. Meeting FERPA's consent, school official exception, and recordkeeping obligations for AI deployments depends on runtime controls, including distinct agent identity, least-privilege access to record fields, and auditable logs of every tool call, not policy or contract language alone.
FERPA Was Not Written for AI Agents
FERPA's implementing regulations (34 CFR Part 99) were written around human school officials requesting or receiving education records, not around AI agents that query a student information system, learning management system, or advising platform on a recurring, automated basis. Applying FERPA to an AI deployment means treating each agent query against a student record as a disclosure event, and evaluating that event against the same consent requirements, school official exception, and legitimate educational interest standard that would apply to a human employee performing the same lookup.
That framing changes the compliance question from whether a vendor contract covers AI use, to whether a specific agent, with a specific scope of access, meets FERPA's disclosure conditions at the moment it runs.
Governance Decisions Above the Technical Layer
Policy and contract language establish who is authorized to access student education records and for what purpose. On their own, they do not enforce those boundaries at the moment an AI agent executes a query. That enforcement happens at the runtime layer: through the agent's identity, the permissions scoped to its task, and the log entry created when it calls a tool. The controls below describe what that runtime layer needs to include to support FERPA's consent, disclosure, and recordkeeping requirements.
AI Runtime Controls That Support FERPA Obligations
- 1
Distinct machine identity for AI agents
Separate agent credentials from human user accounts so access to student records can be individually scoped, logged, and revoked.
- 2
Field-level, least-privilege permissions
Grant agents access to specific record fields required for a task rather than broad SIS, LMS, or advising platform access.
- 3
Tool-call audit logging
Log each agent's queries against student records at the individual tool-call level, not just at session or application level.
- 4
Segregated directory vs. non-directory access paths
Route directory information and non-directory education records through separate access controls to preserve FERPA's differing consent requirements.
- 5
Data minimization at the retrieval layer
Return only the fields an agent needs for a given task rather than passing full student records into agent context.
FERPA Obligations Mapped to AI Runtime Controls
The table below maps core FERPA obligations to the runtime control that addresses them in an AI deployment.
| FERPA Obligation | Runtime Control |
|---|---|
| Consent & Disclosure Limits | Distinct AI agent identity with scoped access permissions |
| School Official Exception | Documented, technically enforced vendor direct control |
| Legitimate Educational Interest | Role-based, field-level access rules for agent functions |
| Recordkeeping (34 CFR 99.32) | Tool-call-level audit trails of agent queries |
Frequently Asked Questions
Does contracting with an AI vendor automatically satisfy FERPA's school official exception?
No. The exception requires the institution to maintain direct control over the vendor's use and retention of education records and to limit that use to the disclosed purpose. This must be established and documented, not assumed from standard vendor terms.
How is directory information different from other education records when an AI tool is involved?
Directory information may be disclosed without consent if notice and opt-out procedures exist. AI outputs that combine directory and non-directory fields can create re-identifiable information the directory exception does not cover, so access paths should be segregated.
What should a FERPA risk assessment for an AI tool cover?
At minimum, it should document which education record fields the tool can access, how that access is authenticated, how disclosures are logged, and which FERPA disclosure basis applies to the use case.
How should AI agent audit logs differ from existing human user access logs?
They need to capture agent identity, the specific record fields accessed, the policy under which access occurred, and the stated purpose, at the level of each individual tool call rather than a session-level login record.
Evaluate AI Deployments Against FERPA-Aligned Runtime Controls
Trussed AI provides runtime governance for AI agents, including agent identity, least-privilege permissions, tool approval workflows, and audit logging that support FERPA-aligned access controls for AI systems handling student education records.
Talk to an Expert