--- title: "Is Claude FERPA Compliant? A 2026 Guide for Schools and EdTech" description: "Claude is not FERPA compliant by default — no AI is. Consumer Claude.ai accounts cannot hold education records; Claude for Work and the API can sit inside a…" url: "https://sonomos.ai/blog/is-claude-ferpa-compliant-2026/" source: "https://sonomos.ai/blog/is-claude-ferpa-compliant-2026/" --- [Back to Blog](https://sonomos.ai/blog/) September 1, 202614 min read Claude FERPA Education AI Compliance Student Privacy # Is Claude FERPA Compliant? A 2026 Guide for Schools and EdTech S [Sonomos Research](https://sonomos.ai/team/) · Sonomos The Sonomos research team writes about AI privacy, data protection, and how to use generative AI safely at work. **Short answer:** Claude is not "[FERPA](https://sonomos.ai/glossary/ferpa/) compliant" out of the box — and no AI product is, because FERPA compliance is a posture the educational institution assembles, not a label a vendor can apply. Claude.ai consumer accounts (Free and Pro) are unsuitable for education records: they are individual consumer contracts with no institutional agreement, so they cannot satisfy the school-official exception in 34 CFR § 99.31(a)(1). Claude for Work (Team and Enterprise) and the Claude API _can_ sit inside a FERPA-compliant deployment — but only if the district or institution signs a written agreement that establishes direct control, restricts re-disclosure, and excludes student data from training. Unlike HIPAA, FERPA has no standard "BAA" instrument; the contract itself does all the work. This guide runs the analysis Claude-specifically, with a brief comparison to ChatGPT and Gemini. For the schools-wide framework covering all vendors, COPPA, and state student-privacy laws, see our companion guide, [FERPA and AI: Can Schools and EdTech Use ChatGPT With Student Data?](https://sonomos.ai/blog/ferpa-ai-chatgpt-claude-schools-edtech-2026/). ## What FERPA actually restricts The Family Educational Rights and Privacy Act (20 U.S.C. § 1232g; 34 CFR Part 99) applies to educational agencies and institutions receiving federal Department of Education funding — effectively all public K-12 districts and nearly all colleges. It is a "no disclosure without consent" statute: **education records**, and **personally identifiable information (PII) from education records**, may not be disclosed to third parties without prior written consent from the parent or eligible student, unless an enumerated exception applies. Three definitions decide most AI questions: - **Education records** — any records directly related to a student and maintained by the institution or a party acting for it: grades, IEPs, 504 plans, disciplinary files, advising notes, and the AI chat transcripts themselves once they contain student information. - **PII from education records** — direct identifiers (name, student ID), indirect identifiers (date of birth), and any other information that would let a reasonable person in the school community identify the student. - **Directory information** (§ 99.37) — a narrow opt-out category that does not cover grades, services, or behavior. A prompt pairing a student's name with a benchmark score is an education-record disclosure, not a directory disclosure. The moment a teacher's prompt containing that PII reaches Anthropic's servers, a disclosure has occurred. Whether it was _permitted_ depends entirely on what sits between the institution and Anthropic contractually. The [schools guide](https://sonomos.ai/blog/ferpa-ai-chatgpt-claude-schools-edtech-2026/) covers the definitions and exceptions in depth; this post applies them to Claude. ## The school-official exception, applied to Anthropic There is no realistic consent path for routine AI use — § 99.30 consent must be signed, dated, and specific to the records, purpose, and recipient. The operative exception is **school official with legitimate educational interest** under § 99.31(a)(1). An outside vendor such as Anthropic can be a school official only if four conditions hold: 1. Anthropic performs an institutional service or function for which the school would otherwise use employees; 2. Anthropic is under the **direct control** of the institution with respect to the use and maintenance of education records; 3. Anthropic is contractually subject to FERPA's restrictions on **further use and re-disclosure** of PII; and 4. The institution uses **reasonable methods** to limit access to records in which there is a legitimate educational interest. Run Claude's tiers through that test: - **Claude.ai Free and Pro fail all four.** They are consumer subscriptions between Anthropic and an individual. The institution is not a party, has no direct control over how prompts are handled, and has imposed no re-disclosure restrictions. On the Free tier, conversations may be used for model training unless the user opts out — the opposite of direct control. Pro excludes training by default, a better posture than some consumer competitors, but a favorable training default does not create a school-official relationship. The analysis is structurally identical to why consumer tiers fail under [GDPR](https://sonomos.ai/blog/is-claude-gdpr-compliant-2026/) and [HIPAA](https://sonomos.ai/blog/is-claude-hipaa-compliant-2026/): the missing artefact is the institutional contract. - **Claude for Work (Team and Enterprise) and the Claude API can pass** — if, and only if, the institution's executed agreement actually establishes the four conditions. Anthropic's business tiers come with a Data Processing Addendum, exclude prompts and outputs from training by default, and (on Enterprise) add SSO, admin controls, and audit logs that support the "reasonable methods" prong. But the DPA is a general-purpose data-protection instrument; districts typically need FERPA-specific school-official designation language and, in many states, a student-data addendum on top of it. One point worth stating plainly: **"FERPA certified" does not exist.** There is no Department of Education certification program, no FERPA seal, and no regulator-blessed equivalent of a HIPAA Business Associate Agreement. Any vendor marketing implying otherwise is describing its own contract terms, not a government designation. The written agreement between institution and vendor is the compliance artefact — full stop. ## Which Claude tier can sit inside a FERPA-compliant deployment? | Tier | Institutional contract possible? | Default training | Can satisfy § 99.31(a)(1)? | Notes | | --- | --- | --- | --- | --- | | Claude.ai Free | No — consumer terms | May train unless user opts out | No | Never suitable for education records | | Claude.ai Pro | No — consumer terms | Excluded by default | No | Better training default; still no institutional agreement | | Claude for Work (Team) | Yes — Anthropic DPA | Excluded by default | Contract-dependent | Needs FERPA school-official language and any state addenda | | Claude for Work (Enterprise) | Yes — Anthropic Enterprise DPA | Excluded by default | Contract-dependent | SSO, admin controls, audit logs support the "reasonable methods" prong | | Claude API | Yes — Anthropic DPA | Excluded by default | Contract-dependent | Zero-data-retention option for eligible customers; common EdTech integration path | "Contract-dependent" is doing real work in that table. A district that provisions Claude Team seats but never executes an agreement with school-official designation language has bought better defaults, not compliance. ## What to verify in Anthropic's terms before an education deployment FERPA does not prescribe a contract form, so procurement teams have to build the checklist themselves. For a Claude deployment, verify: - **School-official designation language.** The agreement should state that Anthropic performs an institutional function, acknowledge direct control over education records, and restrict use of student PII to providing the contracted service. Anthropic's standard [DPA](https://sonomos.ai/blog/claude-dpa-explained-2026/) addresses processing on the customer's instructions; confirm it — plus any education-specific terms your account team offers (the schools guide notes Claude for Education / Work among the tiers with suitable terms) — maps to the four § 99.31(a)(1) conditions. Verify the current offering directly rather than relying on a blog post's snapshot. - **Training exclusion in writing.** Business-tier prompts are excluded from training by default; the contract should say so, because "further use" of student PII for model improvement is exactly what the exception prohibits. - **Retention and deletion.** Confirm retention windows, deletion on termination, and — for API workloads carrying student data — whether the zero-data-retention option applies to your use case. Direct control is hard to square with indefinite vendor retention. - **Sub-processor disclosure and breach notification.** Anthropic publishes a sub-processor list for DPA customers. Map breach-notification timelines to the strictest applicable state student-privacy law (Illinois SOPPA's 30 days, for example). - **State-law addenda.** New York Education Law § 2-d, California AB 1584/SOPIPA, Illinois SOPPA, and peers impose their own contract requirements on top of FERPA. Most districts maintain a standard student-data addendum; Anthropic's willingness to sign it is a procurement question to resolve before rollout, not after. - **Audit and records-access support.** Parents and eligible students have inspection rights (§ 99.10), and AI transcripts containing student information are themselves education records. Enterprise audit logs help the institution answer what was disclosed, by whom, and when. ## COPPA and students under 13 For K-12 districts, [COPPA](https://sonomos.ai/glossary/coppa/) stacks on top of FERPA whenever an online service collects personal information from children under 13. The FTC's school-consent pathway lets a school consent on parents' behalf — but only where the operator collects the data solely for the school's use and benefit and the school keeps control. Consumer Claude.ai does not fit that pathway, and — like other consumer AI chatbots — Anthropic's consumer terms impose minimum-age restrictions that rule out direct under-13 use regardless of parental permission; check the current Acceptable Use Policy for the exact lines. The practical consequence: any Claude-powered experience that under-13 students touch should reach them only through an institutionally contracted deployment (or an EdTech product built on the Claude API) whose terms are designed for the under-13, school-consent context — never through students signing into claude.ai. The [schools guide](https://sonomos.ai/blog/ferpa-ai-chatgpt-claude-schools-edtech-2026/) covers the COPPA analysis and state overlays like SOPIPA and § 2-d in detail. ## How ChatGPT and Gemini compare on FERPA The legal test is vendor-neutral, so the comparison is about contracting paths and defaults, not about any model's intrinsic "compliance": - **ChatGPT.** Consumer tiers (Free, Plus, Pro) fail the school-official test for the same structural reason — no institutional contract — with the added wrinkle that consumer ChatGPT trains on conversations unless the user opts out. ChatGPT Enterprise and ChatGPT Edu (OpenAI's higher-education tier) offer FERPA-aligned terms under a properly executed agreement. See [Is ChatGPT GDPR Compliant?](https://sonomos.ai/blog/is-chatgpt-gdpr-compliant-2026/) for the adjacent tier analysis. - **Gemini.** The consumer Gemini app is the wrong vehicle for student data, but districts already inside Google Workspace for Education can enable Gemini under existing Workspace education terms, which are generally FERPA-aligned and familiar to K-12 procurement. The caveat: older signed addenda may predate the AI features — confirm they cover Gemini specifically. See [Is Gemini GDPR Compliant?](https://sonomos.ai/blog/is-gemini-gdpr-compliant-2026/) for Google's tier posture. - **Claude.** No consumer path, same as the others; the business tiers' distinguishing features are training exclusion by default on every paid tier and availability through AWS Bedrock and Google Vertex AI, letting districts and EdTech vendors inherit an existing cloud agreement's contractual framework rather than negotiating from scratch. In every case the deciding artefact is identical: a written agreement that makes the vendor a school official. The [schools-wide guide](https://sonomos.ai/blog/ferpa-ai-chatgpt-claude-schools-edtech-2026/) compares all four major vendors, including Microsoft Copilot for Education. ## Where Claude deployments go wrong in schools The generic failure patterns — pasting an IEP into a consumer chatbot, drafting discipline write-ups with names — are covered in the schools guide. Three Claude-specific variants show up in practice: 1. **The personal Pro account that "doesn't train."** A special-education teacher uses their own Claude.ai Pro subscription to reword an IEP goal, reasoning that Claude doesn't train on Pro conversations so "nothing bad happens to the data." The training default is real, but irrelevant: the disclosure to a non-contracted third party is the FERPA violation, complete the moment the prompt is sent. This is the education version of the shadow-AI problem in our [HIPAA guide](https://sonomos.ai/blog/is-claude-hipaa-compliant-2026/) — the better consumer default makes the rationalization easier, not the disclosure lawful. 2. **The identical login screen.** A district licenses Claude for Work, but staff who already had personal Claude.ai accounts keep using them on work devices — the chat interface looks the same either way. Grading support with student names flows into the uncontracted account. The fix is identity- and network-layer: SSO-only access to the managed workspace, and consumer endpoints blocked on district devices. 3. **The EdTech pass-through.** A district's LMS or tutoring vendor adds an assistant feature built on the Claude API. The district's contract with the EdTech vendor has school-official language — but the vendor's downstream terms with Anthropic may or may not pass FERPA obligations through, or have zero retention configured. The institution owns the whole chain; ask the vendor for its sub-processor terms. ## Where Locke fits The strongest position under FERPA is the one where no PII from education records leaves the device in the first place. FERPA itself points this direction: § 99.31(b) permits releasing records without consent after removal of all personally identifiable information, provided the institution reasonably determines the student is not identifiable, taking into account other reasonably available information. That is a genuine legal standard, not a checkbox: quasi-identifiers ("the only 7th-grader at Lincoln with a one-to-one aide") can identify a student with no name present. Our [de-identification](https://sonomos.ai/glossary/de-identification/) glossary entry covers why this is harder than it looks. Locke supports that posture as a technical control. It detects student names, ID numbers, dates of birth, and similar education-record signals locally in the browser and replaces them with [pseudonymised](https://sonomos.ai/glossary/pseudonymization/) placeholders before the prompt leaves the device — whether the destination is a sanctioned Claude Enterprise workspace or a personal account the policy failed to catch. Because detection runs entirely on-device and Sonomos never receives the text, installing Locke does not create a new vendor disclosure needing its own FERPA agreement. To be precise about the claim: Locke reduces the PII that reaches any AI endpoint and supports a de-identification workflow; it does not by itself render a dataset de-identified under § 99.31(b), and it does not make any Claude tier "FERPA compliant." No tool can. ## Frequently asked questions ### Is Claude FERPA compliant? No AI product is FERPA compliant in itself. Claude.ai Free and Pro cannot be used with education records because no institutional contract exists. Claude for Work and the Claude API can operate inside a FERPA-compliant deployment when the institution executes an agreement establishing the § 99.31(a)(1) school-official conditions — direct control, re-disclosure restrictions, training exclusion, and access limits — plus any state student-data addenda. ### Is there such a thing as a "FERPA compliant AI"? As a product label, no — and there is no FERPA certification a vendor can hold. What exists are AI products whose enterprise or education tiers offer terms that _can_ satisfy the school-official exception when contracted properly: Claude for Work and the API, ChatGPT Enterprise/Edu, Gemini through Workspace for Education, and Copilot through the Microsoft Education framework. The compliance lives in the institution's contract, configuration, and controls. ### Is ChatGPT FERPA compliant? Same structure: consumer ChatGPT (Free, Plus, Pro) cannot satisfy the school-official exception; ChatGPT Enterprise and ChatGPT Edu can under a properly executed agreement with training disabled and admin controls configured. Details in the [schools guide](https://sonomos.ai/blog/ferpa-ai-chatgpt-claude-schools-edtech-2026/). ### Is Gemini FERPA compliant? The consumer Gemini app is not suitable for education records. Gemini enabled within eligible Google Workspace for Education plans operates under education terms generally aligned with FERPA — but districts should confirm their signed state-law addenda actually cover the Gemini features being turned on. ### What is the most secure AI setup for a K-12 district in 2026? There is no single "most secure AI"; there is a most defensible stack: (1) a contracted enterprise or education tier — Claude for Work, ChatGPT Edu/Enterprise, Gemini via Workspace for Education, or Copilot for Education — with FERPA school-official language, training excluded, and your state's student-data addendum signed; (2) vendor attestations reviewed ([SOC 2 Type II](https://sonomos.ai/blog/soc-2-ai-chatgpt-claude-2026/) is table stakes in RFPs); (3) COPPA's school-consent conditions met for any under-13 exposure; (4) consumer AI endpoints blocked on managed devices; and (5) a local redaction layer masking student PII before any prompt leaves the browser. Rank vendors on willingness to sign your terms, not on marketing language. ### Can teachers use Claude for lesson planning without a district contract? Yes, for genuinely student-free work. "Draft a lab on photosynthesis for 9th grade" implicates no education record. The line is crossed when the prompt describes an identifiable student — by name or by combination of details. District policy should draw that line explicitly rather than banning or blessing the tool wholesale. ## The bottom line "Is Claude FERPA compliant?" resolves to three narrower questions: _which Claude am I using_, _does my institution have a written agreement that makes Anthropic a school official for that product_, and _can we keep staff and students inside that contracted boundary?_ Claude's business tiers give a district good raw material — training excluded by default, a DPA, enterprise controls, cloud-marketplace contracting paths — but the institution assembles the compliance: the contract, the state addenda, the configuration, the workforce controls, and a redaction layer for the day policy meets a deadline. ## Related reading - [FERPA and AI: Can Schools and EdTech Use ChatGPT With Student Data? (2026 Guide)](https://sonomos.ai/blog/ferpa-ai-chatgpt-claude-schools-edtech-2026/) - [Is Claude GDPR Compliant?](https://sonomos.ai/blog/is-claude-gdpr-compliant-2026/) - [Is Claude HIPAA Compliant?](https://sonomos.ai/blog/is-claude-hipaa-compliant-2026/) - [Claude's DPA, Explained](https://sonomos.ai/blog/claude-dpa-explained-2026/) - [Is ChatGPT GDPR Compliant?](https://sonomos.ai/blog/is-chatgpt-gdpr-compliant-2026/) - [Is Gemini GDPR Compliant?](https://sonomos.ai/blog/is-gemini-gdpr-compliant-2026/) - [SOC 2 and AI Tools](https://sonomos.ai/blog/soc-2-ai-chatgpt-claude-2026/) ### Protect your data while using AI Locke detects and masks sensitive information on your device, before it reaches an AI model. [Download Locke](https://sonomos.ai/downloads/) This article is general information about privacy and AI, not legal advice, and reading it does not create a solicitor–client or attorney–client relationship. Regulatory requirements turn on your specific facts and jurisdiction and change over time. Consult qualified counsel before relying on anything here, and check the publication and last-reviewed dates above. We may revise posts, but we do not undertake to keep any post current as the law changes. ## Reference pages for this post - [Locke for ClaudeTool guide](https://sonomos.ai/for/claude/) - [FERPA (Family Educational Rights and Privacy Act)Glossary](https://sonomos.ai/glossary/ferpa/) ## Related Articles September 1, 2026•Sonomos Research ### The Anthropic DPA Explained: Claude Data Processing Agreement Guide (2026) Anthropic offers a DPA on Claude Team, Enterprise, and the API — not on Claude.ai Free or Pro. Here is what the Anthropic DPA covers, how it handles EU-US transfers via SCCs and the DPF, what to verify before signing, and where a DPA stops. Claude DPA GDPR +2 September 1, 2026•Sonomos Research ### Claude EU Data Residency in 2026: What Anthropic Offers, and What European Teams Do Instead Anthropic does not currently offer EU data residency for Claude — every prompt is processed on US infrastructure and constitutes an EEA-to-US transfer. Here is how that compares to OpenAI, Microsoft, and Google, what the EU-US Data Privacy Framework does and does not cover, and the supplementary-measure stack European teams deploy instead. Claude Data Residency Europe +2 September 1, 2026•Sonomos Research ### Can My Employer See My Claude Chats? What's Visible in 2026 (and What Isn't) Whether your employer can see your Claude chats depends on the account (personal Claude.ai vs. work-managed Team/Enterprise), the device, and the network. Here is what is visible in each scenario — including Claude Code — and what deleting a chat actually does. Claude Workplace Privacy AI Privacy +1