Student Information Systems With AI Features and the FERPA Audit Question

Student Information Systems With AI Features and the FERPA Audit Question

SIS vendors are in an arms race to bolt AI features onto their platforms. PowerSchool, Infinite Campus, Ellucian, Anthology, and a growing list of others have all announced or shipped AI capabilities in the last 18 months. These range from predictive analytics for student retention to AI-generated progress reports to chatbots that pull from student records to answer parent and advisor queries. Some of this is genuinely useful. Some of it is a checkbox on an RFP response. But all of it raises a FERPA question that most districts and universities have not seriously worked through.

The question is not whether AI features in an SIS can comply with FERPA. They can. The question is whether your institution has the audit posture to demonstrate that compliance if the Department of Education's Student Privacy Policy Office (SPPO) comes asking, or if a parent files a complaint, or if a breach forces disclosure of how student data was actually flowing.

The School Official Exception Is Doing a Lot of Heavy Lifting

Most SIS vendors access student education records under the "school official" exception in 34 CFR 99.31(a)(1). This is the provision that lets institutions disclose personally identifiable information (PII) from education records to contractors, consultants, and other parties performing institutional services, without obtaining prior written consent from parents or eligible students. The vendor has to be under the "direct control" of the institution with respect to the use and maintenance of education records, and the institution's annual FERPA notification has to specify criteria for who qualifies as a school official.

This exception works fine for a traditional SIS doing traditional SIS things: storing grades, managing enrollment, generating transcripts. The data flows are well understood. The vendor's use of PII is bounded by the service agreement. Auditors can trace what data goes where and why.

Now add an AI feature that ingests student records to train or fine-tune a model. Or one that sends student data to a third-party LLM API for inference. Or a predictive analytics module that combines education records with behavioral data from an LMS. The school official exception still applies in theory, but the "direct control" and "legitimate educational interest" requirements become much harder to document and verify.

Where the Gaps Show Up

Here is what I keep seeing when institutions try to audit their AI-enabled SIS deployments:

• Subprocessor opacity. The SIS vendor is your school official. But if that vendor sends student PII to OpenAI, Anthropic, Google, or AWS Bedrock for inference, those subprocessors are not named in your school official designation. SPPO's 2023 guidance on third-party service providers makes clear that the institution remains responsible for ensuring all downstream recipients meet school official criteria. Most institutions have no visibility into these subprocessor relationships, let alone contractual controls over them.
• Model training ambiguity. Does the AI feature use student data to improve the model? If so, is that a "legitimate educational interest" for your institution, or is it product development for the vendor? The distinction matters enormously under FERPA. The 2011 FERPA final rule (76 FR 75604) narrowed the school official exception specifically to prevent vendors from using education records for purposes other than the contracted service. If student data is training a model that benefits other customers, you have a problem.
• De-identification claims that do not hold up. Vendors frequently argue that data sent to AI services is de-identified and therefore outside FERPA's scope. Under 34 CFR 99.31(b), properly de-identified data requires removal of all reasonable identifiers and application of methods that a "reasonable person" in the school community could not use to re-identify students. With modern LLMs processing rich contextual data (narrative progress notes, behavioral incident descriptions, accommodation details), the re-identification risk is real, and few vendors have subjected their de-identification methods to serious scrutiny.
• Audit trail gaps. FERPA requires institutions to maintain a record of each disclosure of PII from education records (34 CFR 99.32). When an AI chatbot pulls from a student's record to generate a response, is that a disclosure? To whom? Is it logged? Most AI features in SIS platforms generate no audit trail that maps to FERPA's disclosure logging requirements.

What a Defensible Audit Posture Looks Like

The SPPO has been increasingly active. In fiscal year 2023, the office investigated over 2,000 complaints, and its enforcement letters have started referencing cloud services and third-party data sharing with more specificity. The Edmodo breach in 2023, which exposed data on 77 million users, and the PowerSchool breach disclosed in January 2025, which affected districts across North America, have put vendor security and data handling practices under a spotlight. Institutions that cannot demonstrate how they maintained direct control over education records in AI-enabled vendor systems are exposed.

A defensible posture requires several concrete things:

• Contractual specificity on AI data flows. Your data processing agreement with the SIS vendor needs to explicitly address whether student PII is used for model training, whether it is sent to subprocessors for inference, and what de-identification methods are applied. Generic "we comply with FERPA" language in a vendor contract is worth nothing in an audit.
• Subprocessor mapping. You need a current list of every entity that touches student education records, including AI inference providers. This should be an exhibit to your DPA, updated at least annually, with notification requirements for changes.
• Technical controls on AI features. Can you disable specific AI features that you have not vetted? Can you limit which student populations are included in AI processing? If the vendor does not offer granular controls, you are accepting risk you may not have assessed.
• Disclosure logging for AI interactions. Work with your vendor to ensure that AI-generated outputs derived from education records are logged in a way that satisfies 34 CFR 99.32. This is an area where most platforms are behind, and your asking about it in procurement or renewal conversations will move the needle.
• Annual review tied to FERPA notification. Your annual FERPA notification defines who qualifies as a school official. If your SIS vendor has added AI features since your last notification, review whether your criteria still accurately describe the vendor's role and access. This is an easy thing to let slip, and it is one of the first things SPPO checks.

The Records Access and Amendment Angle

One more thing worth flagging. Under 34 CFR 99.10 and 99.20, parents and eligible students have the right to inspect education records and request amendments. If an AI feature generates a prediction, risk score, or narrative summary that becomes part of a student's education record, the institution needs to be prepared to explain what data informed that output and to process amendment requests. The Family Policy Compliance Office (now SPPO) addressed a version of this in its 2020 guidance on automated decision-making in education, noting that algorithmically generated records are still education records if they are directly related to a student and maintained by the institution. Districts using AI-generated early warning indicators or behavioral risk scores should think carefully about whether those outputs are being maintained in a way that triggers these rights.

How FirmAdapt Addresses This

FirmAdapt's architecture was built around the assumption that regulated organizations need to maintain auditable control over how AI systems interact with protected data. For education institutions, this means FirmAdapt can map data flows between your SIS, any AI processing layers, and downstream subprocessors, then generate the documentation you need to demonstrate FERPA compliance at the disclosure, access, and de-identification levels. The platform maintains granular audit logs that align with 34 CFR 99.32 requirements, even when AI features are generating outputs derived from education records.

FirmAdapt also supports ongoing contract and policy review by flagging when vendor AI capabilities change in ways that affect your school official designations or your annual FERPA notification language. If your SIS vendor adds a new AI feature that routes student data through a subprocessor you have not vetted, FirmAdapt surfaces that gap before it becomes an audit finding.