AI in the Classroom: Are Vendors Actually "School Officials" Under FERPA?

Why Consumer AI Violates the Exception

Many educators assume that using consumer-facing tools like the free tier of ChatGPT or generic chatbots is harmless. In reality, these tools often fall outside the "School Official" framework, leaving the disclosure without a lawful basis. Tested against the four criteria above, a consumer tool usually fails each one.

Outsourced function: A teacher who independently signs up for a chatbot has not been asked to do so by the school, so there is no function that has been formally outsourced to the AI platform by the school.

Direct control: The school cannot dictate retention, deletion, or security on a platform governed by public terms of service that the vendor is free to change at any time.

Strict use limitations: Most general-purpose models retain inputs and reuse them to train future commercial products, a secondary use the school never authorized.

Legitimate educational interest and data minimization: An open text box will accept whatever the user chooses to input, including a full IEP or behavioral history. It should be limited to a narrow set of fields that the task actually requires.

Failing any single criterion is enough to take the disclosure outside the School Official exception. Uploading an education record to a consumer platform can therefore result in an unauthorized disclosure and a secondary use of data, which violates the redisclosure and use limitations of FERPA.

"No Training" is Not Enough

Using a tool as a compliant school official requires planning. Simply agreeing to a "we do not train models on school data" clause in a terms-of-service page is insufficient, particularly where the vendor reserves the right to amend those terms unilaterally.

True FERPA compliance for an AI platform requires an explicit Data Privacy Agreement (DPA) that handles the realities of AI architecture:

• AI tools must restrict data processing strictly to the contracted scope (e.g., generating a grade or providing feedback). Once the task is completed, data retention and deletion pipelines must be clearly defined.
• AI platforms frequently rely on secondary APIs, hosting providers, or other external systems. If a vendor passes student data to a subprocessor (such as an external Large Language Model (LLM) provider), that subprocessor must be bound by the same rules the school imposed on the vendor.
• In addition to traditional student data (such as grades or assignments), predictive analytics, AI-generated behavioral scores, or saved chatbot conversation logs involving mental health support are legally considered education records. They are subject to the same student access, review, and amendment rights as a traditional report card. The DPA must also require that the vendor support those rights operationally, including retrieval and correction of AI-generated outputs.

FERPA is also only the floor. State student-privacy statutes frequently impose stricter contractual requirements and should be evaluated when determining what AI tools can be used by educators with student data. For example, New York Education Law § 2-d and 8 NYCRR Part 121 require a signed data privacy agreement, a Parents' Bill of Rights, and a data security plan for every contract involving student personally identifiable information. California's SOPIPA (Bus. & Prof. Code § 22584) regulates the vendor directly, barring targeted advertising, non-educational profiling, and any sale of student data. Illinois' SOPPA (105 ILCS 85) requires public posting of agreements. Finally, Colorado, Connecticut, Texas, Virginia, Florida, and Louisiana impose their own contracting and secondary-use restrictions.

How Educators Can Protect Themselves

School districts and educational institutions must avoid relying on ad hoc, unvetted AI adoption. They must establish formal data governance policies with respect to the adoption and use of AI tools that: (1) inventory the AI tools already in use across classrooms, including free and teacher-adopted tools; (2) mandate the use of privacy-first platforms that have signed DPAs designating the vendor as a School Official; (3) require vendors to disclose subprocessors and to flow-down all contractual obligations between vendor and the school to any downstream subprocessor/model provider; and (4) provide ongoing training to educators, administrators, and staff on the permissible uses of AI, applicable laws, and the institution's AI governance and acceptable use policies.

AI will remain in classrooms whether or not school districts plan for it, and the School Official Exception is wide enough to accommodate it, but only if properly documented in agreements that can constrain the vendor. The distinction that matters is not whether a tool is "AI," but whether the educational institution retains direct control over education records throughout the processing chain, including downstream model providers and subprocessors.