Is ChatGPT FERPA Compliant? The Honest Answer for Schools

Is ChatGPT FERPA compliant?

No — and neither is any other piece of software, because "FERPA compliant" isn't a property a tool can have on its own. FERPA regulates schools. ChatGPT becomes part of a compliant deployment only when a district signs an agreement placing an education or business tier under FERPA's school-official exception.

That reads like a lawyer's dodge, so it's worth saying plainly what it means in practice. When a vendor's sales page says "FERPA compliant," it is making a claim the law doesn't recognize; when OpenAI's own materials say ChatGPT for Teachers is "built to protect student data and help schools meet FERPA requirements," that careful phrasing is actually the legally accurate one. The compliance question always decomposes into three smaller questions: which tier of the product is your school using, what agreement exists between your district and the vendor, and what are you personally typing into it. As of July 2026, ChatGPT's tiers differ sharply on all three — enough that the same prompt can be routine on one account and a reportable data incident on another.

This page walks the tiers. The legal machinery underneath — what FERPA's definitions of education records and personally identifiable information actually cover, what a data privacy agreement does, and five worked paste-or-don't-paste scenarios — lives in our student data privacy guide, and this page leans on it rather than repeating it. Standard caveat, sincerely meant: this is practical guidance from primary sources, not legal advice, and your district's counsel and policy outrank anything written here.

Why "compliant" belongs to the deployment, not the product

The Family Educational Rights and Privacy Act binds educational institutions that receive federal funds. It restricts how schools disclose personally identifiable information from education records; a software vendor is never directly subject to it. So when a district wants an outside tool to touch student data legitimately, it almost always routes through the school-official exception, which lets the district treat a contractor as an extension of its own staff.

The U.S. Department of Education's guidance on online educational services (February 2014, still the operative framework) sets the conditions — most decisively, that the provider stay "under the direct control of the school or district with regard to the use and maintenance of education records"; our privacy guide walks through all four. In practice, direct control means a contract — and the same guidance warns districts to be cautious about click-wrap terms "that allow for amendment without notice," precisely because a vendor who can rewrite the deal unilaterally isn't under anyone's direct control.

Hold ChatGPT's tiers up against that framework and the product splits cleanly in two: versions where OpenAI will sign the necessary agreement, and versions where no such agreement is available to you at any price.

The tiers, as of July 2026

OpenAI's lineup shifts often — the Team plan was renamed ChatGPT Business on August 29, 2025, and the education products have been reorganized more than once — so treat this as a snapshot with a date on it, and check the linked pages before your district signs anything.

Consumer ChatGPT (Free, Plus, Pro). Personal accounts under OpenAI's consumer terms. Conversations can be used to train OpenAI's models by default; an individual can opt out, and Temporary Chats aren't used for training, but there is no version of this tier a district can contract for. This is the account a teacher typically already has.

ChatGPT Business (formerly Team). A paid workspace tier. Per OpenAI's enterprise privacy commitments, business-tier data isn't used for training by default, workspace admins control retention and can view, export, and delete user conversations, and OpenAI executes a Data Processing Addendum with customers. Built for companies rather than schools — it lacks the student-data-specific contract language a district needs, though the underlying data handling is the same class as the education tiers.

ChatGPT Enterprise and ChatGPT Edu. The institutional tiers. ChatGPT Edu is "built for universities to responsibly deploy AI to students, faculty, researchers, and campus operations" — with no training on institutional data, custom retention windows, SAML SSO and SCIM provisioning, AES-256 encryption at rest, and SOC 2 Type 2 certification. The Enterprise and Edu tiers include a Compliance API for auditing conversations. These are the tiers where OpenAI's student-data paperwork applies.

ChatGPT for Teachers. The K-12 offering: a workspace free for verified U.S. K-12 teachers, staff, and administrators through June 2027, with identity verification through SheerID, district domain claiming, role-based access, and SSO. OpenAI's help documentation states that workspace content "is not used to train our models by default" and that the plan carries "FERPA-aligned protections and a student data privacy agreement." It also states, in exactly these words: "This plan is not for students." It's a staff-side tool — teachers planning, drafting, and analyzing — not a student-facing one.

Why your personal account can't be fixed with a settings toggle

A teacher who has read this far and turned off "Improve the model for everyone" in their personal account's Data Controls has done something worthwhile — and has not moved one inch toward FERPA coverage. The training toggle addresses one specific risk. The compliance gap is structural, and it has three parts.

First, the missing agreement. When you created your account, you accepted OpenAI's consumer terms of use alone, as an individual. Your district is not a party to that contract, cannot enforce it, cannot audit under it, and cannot compel deletion through it. Whatever the settings say, there is no mechanism by which student data in your personal account is "under the direct control of the school or district." That's the condition the whole school-official exception hinges on, and no checkbox creates it.

Second, retention you don't control. OpenAI's stated policy is that deleted conversations and Temporary Chats are removed from its systems within 30 days — but 2025 provided a live demonstration of the limits of that promise on consumer accounts. During The New York Times' copyright litigation, a court preservation order required OpenAI to retain consumer ChatGPT and API content that would otherwise have been deleted, including chats users had deliberately erased. The obligation ran until September 26, 2025, and a locked-down slice of April–September 2025 user data remains preserved for the litigation. Two details matter for schools. The order swept in Free, Plus, Pro, and Team accounts. It excluded ChatGPT Enterprise, ChatGPT Edu, and API customers with zero-data-retention agreements — the tiers where customers had negotiated contractual data commitments. The accounts with signed agreements were the ones a federal court left alone. It's hard to imagine a cleaner argument for why the signature is the substance.

Third, the terms themselves. OpenAI's consumer terms of use require users to be "at least 13 years old or the minimum age required in your country to consent," and anyone under 18 must have a parent or guardian's permission. A teacher directing students to personal ChatGPT accounts is walking into age-gating and consent questions on top of the FERPA ones — and for students under 13, into COPPA territory, which the privacy guide covers in the depth it deserves.

None of this makes a personal account useless. It makes it a tool with a boundary: nothing that identifies a student goes in, ever, on any setting.

If you've already done this — pasted student work or student details into a personal account before you knew the line — the sequence is short. Stop adding to it. Delete the conversations, knowing what the preservation-order episode above showed: on a consumer account, deletion is a request to the vendor, not a guarantee. Turn off model training in Data Controls. Then check whether your district has a data-incident procedure and consider a self-report to whoever runs it — a well-drafted AI policy treats a good-faith self-report as professional learning rather than misconduct, and our school AI policy template writes that amnesty in explicitly.

What OpenAI will actually sign

Here the news is genuinely better than most teachers assume. OpenAI publishes a standing Student Data Privacy Agreement covering its education services, and its language maps directly onto the federal framework. The agreement states that "in performing the Services, OpenAI shall be considered a School Official with a legitimate educational interest" and that "OpenAI shall be under the direct control and supervision of Customer with respect to its use of Student Data." Student data "is and remains the property and under the control of Customer." Subprocessors must be bound to terms no less stringent. If a parent contacts OpenAI to see their child's data, OpenAI refers them to the school; if law enforcement requests student data, OpenAI notifies the district first unless legally barred.

The education terms for the Teachers workspace carry the same spine — "Student Data belongs to the School and remains under the School's control" — plus operational rules worth knowing: the accepting teacher must be at least 18 and authorized by their school, and the account is restricted to educational use in a teaching role (the terms name lesson planning, classroom activities, assessment, and family communication as the intended lanes).

Read those documents next to the 2014 federal guidance and the alignment is unmistakable: school-official designation, direct control, purpose limitation, no re-disclosure. This is what it looks like when a vendor drafts to the exception. And notice what the same documents ask of the district in return — the SDPA obligates the customer to comply with privacy laws, handle required notices and consents, and include its school-official criteria in the annual FERPA notification of rights. The agreement is a duet. A district that signs it and never updates its annual notification has left its own half unfinished, which is a useful reminder that "is ChatGPT FERPA compliant?" was always partly a question about the district.

What a legitimate district deployment involves

For a K-12 district that wants ChatGPT available to staff without anyone freelancing on personal accounts, the path as of July 2026 runs roughly like this:

1. Pick the tier that matches the use. Staff-side work (planning, drafting, data analysis with appropriate controls) fits the ChatGPT for Teachers workspace, or Enterprise and Edu for districts that want the Compliance API those tiers include. Anything student-facing is a different project with different consent requirements, and the Teachers workspace explicitly doesn't cover it.
2. Execute the agreement, not just the account. Verify educators, claim the district domain, and put the Student Data Privacy Agreement (or a state-standard DPA, if your state's alliance requires its own form) through whoever signs vendor contracts. What a DPA legally accomplishes, and how to check your state's registry for existing agreements, is covered in the privacy guide's DPA section.
3. Do the district's half. Update the annual FERPA notification's school-official criteria, set retention windows in the admin console, wire up SSO, and decide who gets workspace access.
4. Tell teachers where the line is. The deployment fails at the last foot if staff don't know that the district workspace and their personal account are legally different places. The guidance that works is concrete: which account for which work, and what still stays out of prompts even inside the covered workspace.

A district can be slow at this for reasons that have nothing to do with hostility to AI — legal review queues, board calendars, a technology director already underwater. If yours hasn't moved, the AI for teachers guide covers how to operate sensibly in the meantime, including what to do when no policy exists at all.

What this means at the teacher level: one worked case

Consider a high-school English teacher with ninety juniors, sitting down in October with a folder of college-essay drafts, wondering whether ChatGPT can help her return feedback faster (the feedback workflow itself — rubric-anchored comment drafting, with the grade staying hers — is in our AI grading and feedback guide).

On her personal Plus account, the honest answer is mostly no. A college essay is the hardest document in a school to de-identify, because it's autobiography by design — the whole genre is a student disclosing the specific texture of their life. Deleting the name from a draft about caring for a younger brother through a parent's deportation case de-identifies nothing; anyone in the building could name that student, and FERPA's definition of personally identifiable information reaches exactly that kind of linkage. She could defensibly paste a generic thesis-statement exercise ("write six sample thesis statements for a college essay about a part-time job, ranging from cliché to sharp, with one line on what separates each from the next") or build a rubric-based feedback prompt with no student text at all. The drafts themselves stay out.

Inside a district ChatGPT for Teachers workspace under the signed agreement, the legal posture changes: student work can enter, because OpenAI is now, contractually, an arm of her school, barred from training on the content or using it for anything but the service. What doesn't change is her judgment about the content itself. District policy may still restrict what goes in; the essay revealing a student's disability status or family legal situation deserves restraint regardless of what the contract permits; and a first-generation senior's college essay is not a worksheet — some feedback should stay fully human because of what the document is, not what the law says. The line between what a covered account permits and what a teacher should still keep in her own hands is its own discussion, and the five scenarios in the privacy guide walk that reasoning step by step.

Her recommendation letters make the same point from the other side. One concession to the rule in our AI for teachers guide — AI gets a rec letter's structure, never the student — still applies: a generic letter is a bad letter, and that was never a legal question. The legal line itself is clean. Drafting a letter with the student's real details in a personal chatbot account would be a disclosure, because once she writes and submits it through school systems, the letter can become part of an education record; drafting the same letter inside the covered workspace is not.

The version to forward

If your principal or tech director asks whether ChatGPT is FERPA compliant, here is an answer you can paste into the email whole:

No software is FERPA compliant by itself — FERPA regulates schools, so the real question is whether we'd be compliant using it. Consumer ChatGPT accounts (Free, Plus, Pro) can't be brought under FERPA's school-official exception at any setting, so nothing student-identifying should ever go into one. ChatGPT for Teachers, Edu, and Enterprise can be covered, because OpenAI publishes a Student Data Privacy Agreement accepting school-official status and district control over student data. Until we've signed one and done our side — annual notification, retention settings, staff guidance — the working rule for everyone is: personal account, no student data, no exceptions.

And once the covered workspace exists, the interesting question stops being legal and becomes practical — what the tool is actually good for on a school day, which is the whole subject of our ChatGPT prompt library for teachers.