FERPA, Cloud Inference, and the Data-Sovereignty Question Procurement Should Be Asking

An accessibility service may process course readings, assessments or student work. Procurement should identify the actual data in those files and every service that receives it. A cloud endpoint is a data flow to review, not evidence of an unlawful disclosure by itself.

Which files are education records?

The Department of Education's FERPA guidance explains that education records are directly related to a student and maintained by the institution or a party acting for it. A generic course reading or blank exam is not automatically an education record. Identifiable graded work or student responses can meet the definition; assess the actual record and applicable exceptions.

If relying on the school-official exception for a contractor, review its conditions, including institutional control over use and maintenance, legitimate educational interests, and permitted use and redisclosure. The Department's online-services guidance is a useful starting point. An executed agreement supports that review; a DPA alone does not establish every FERPA condition.

Trace the complete processing path

Record whether processing is deterministic, uses an institutional model, or calls a vendor or third-party model service. Include storage, logs, backups, support access and telemetry, as well as inference. Provider account settings and applicable terms can differ; do not infer retention or training policy from a provider name or a generic enterprise label.

Ask for written answers:

• What content and metadata leave the institution, for which purpose and destination?
• Which subprocessors receive them, and in which regions?
• What is retained in prompts, outputs, logs and backups, and how is deletion verified?
• Can submitted content be used for training, and which terms and settings govern that use?
• Which agreements cover each recipient, and how are subprocessor changes communicated?
• Who can access the data and investigate an incident?

Self-hosting still requires a data-flow review

Running application containers on institutional infrastructure can provide more operational control. It does not guarantee that data stays there: configured cloud models, integrations, monitoring or support tools can create external flows.

Aelira Core is beta and supports provider configuration and deterministic scanning without a model. Verify the selected release, enabled providers, egress and deployment controls in the public repository. Core is licensed under AGPL-3.0; the CLI has a separate MIT license. Neither license establishes privacy compliance.

For a pilot, use authorized representative inputs, minimize unnecessary student identifiers, inspect outbound connections, and test access, deletion and retention controls. Have the institution's privacy and legal teams assess the actual architecture and agreements before sensitive records enter production. We have not published a representative survey of competitor contracts or higher-education procurement reviews.

Further reading

These links provide additional context; each source should be assessed for its own scope.

AI-Assisted University Accessibility: Review, Evidence and Legal Scope

Resource 2: studentprivacy.ed.gov