Skip to content
Policy and compliance

Anthropic's Binding AI Safety and Privacy Agreement for Schools: What It Covers and What to Ask Any AI Vendor

Elena Volkov · Compliance & Privacy Writer, Borderset · 7 min read

On 10 November 2026, the American Federation of Teachers (AFT), the United Federation of Teachers (UFT) and Anthropic announced a legally binding agreement that applies AI safety and privacy standards to districts using Claude for Teachers. This post explains what the commitments cover, where their limits are, and the questions school leaders should ask any AI vendor before signing.

The short answer

Anthropic has agreed to the AFT's AI Safety and Privacy Standard for Schools, a memorandum of agreement developed through the National Academy for AI Instruction. According to the AFT announcement, the standard applies to Anthropic's existing agreements with school districts using Claude for Teachers. Its core commitments are that student and educator data cannot be used to train AI models (with a narrow safety exception), cannot be sold or used for advertising, stays under the school's control for retention and deletion, and that educators, not AI, make consequential decisions about students. Education Week reports that districts will need to opt in for future licensing agreements, and that the agreement covers only Claude for Teachers. Whatever vendor you use, the same principles make a good checklist. Facts here are as published up to 14 November 2026.

Disclosure: Borderset has no commercial relationship with Anthropic, the AFT or the UFT. This article was prepared with AI assistance and checked against the primary sources linked.

What was announced

ItemDetail
Announced10 November 2026, by the AFT, the UFT and Anthropic
StandardAI Safety and Privacy Standard for Schools, memorandum of agreement, version 1.0 dated September 2026
Developed throughThe National Academy for AI Instruction in New York City, founded by the AFT and UFT with Microsoft, OpenAI and Anthropic
Product coveredClaude for Teachers, launched in July 2026, which gives verified U.S. K-12 educators free access to Claude
Who it applies toDistricts using Claude for Teachers; the AFT says existing agreements are covered automatically
Similar agreementsEducation Week reports Microsoft adopted similar language in September 2026 and that the union is still negotiating with OpenAI

Sources: AFT press release, Education Week, 10 November 2026 and Anthropic's Claude for Teachers announcement. Education Week describes the free access as available for this year at least, and notes that Claude for Teachers is aimed at adult educators and does not include student-facing tools, which makes the agreement narrower in practice than Microsoft's.

What the commitments cover

The AFT summarizes the agreement in three principles: student data is off limits, educators stay in charge, and there are no "black boxes". The full memorandum sets out ten binding principles. Key provisions from the signed text include:

AreaWhat the standard requires
Training on dataNo use of student, educator or school data to train, fine-tune or improve AI models, including de-identified data, with a narrow, auditable exception for safety and security
Data minimizationCollect only what is strictly necessary; no biometric collection, keystroke logging, behavioral tracking or long-term profiling without explicit written approval
Ownership and deletionThe school owns its data, can export it, and controls retention; deletion from active systems within 180 days of a request, with backups purged on the same maximum timeline
Data locationCovered data kept in the European Union or North America, with the location identifiable to the school
MemoryWhere a tool remembers users, they can see, erase or switch off what it stores
Human oversightEducators keep responsibility for consequential decisions; AI disclosure in the product; no companion-style features designed to build emotional dependence
AccountabilityIndependent certifications such as SOC 2 Type II and ISO 27001, 27701 and 42001 (or a dated plan to obtain them); breach notice within 72 hours
TermTwo years from signing, renewable in writing

The memorandum also says participating providers must make its protections available to any U.S. school district that asks, through a simple process, and that the protections can be added to a district's existing data privacy agreement or addendum. Once incorporated, they are enforceable by the district under that contract.

What the agreement does not do

  • It covers one product. Education Week notes it applies to Claude for Teachers, not necessarily to other companies' education tools built on Claude models. Check the terms of each tool you buy.
  • Districts still have work to do. According to Education Week, districts will need to opt in to the standard for future licensing agreements, and the human review commitments depend on districts training staff and setting policy.
  • It does not replace law. The standard says it operates alongside FERPA, COPPA, IDEA and state privacy laws. Where those laws require more, they still apply.
  • Enforcement depends on contracts. The protections become enforceable by a district once they are in that district's agreement. Keep a copy of the signed terms with your vendor records.

The AFT president described the agreement as a step toward an industry standard, while saying the union also wants these protections in law. The UFT president said that, ideally, the federal government would step in.

Questions to ask any AI vendor before signing

The standard is a useful benchmark even if you never use Claude. Ask these questions of every AI tool that will touch student or staff data, and get the answers in writing, ideally in your data privacy agreement.

Data use and training

  1. Will any prompt, file, output or other data from our staff or students be used to train or improve your models, including in de-identified form? What exceptions exist?
  2. Do you sell or share our data with third parties, or use it for advertising?
  3. Which subprocessors handle our data, and where is it stored?

Control, retention and deletion

  1. Who owns the data and outputs created in your product?
  2. Can we set our own retention period and delete data ourselves? How quickly is deleted data removed from active systems and backups?
  3. Can we export all our data in a standard format if we leave?
  4. Does the product remember users between sessions? Can we turn that off?

Safety and oversight

  1. Is the product designed for adults, students, or both? Which age groups and grade levels can we enable or disable?
  2. What safeguards exist for self-harm, violence, sexual content and other high risk topics?
  3. Can the tool take actions on a user's behalf, such as sending messages? Can we disable that for students?
  4. How do you make sure educators, not the AI, make grading, discipline and placement decisions?

Compliance and accountability

  1. How do you support our obligations under FERPA, COPPA and our state's student privacy law?
  2. Which independent security and privacy certifications do you hold, and can we see the reports?
  3. How quickly will you notify us of a breach, and what help will you provide?
  4. Will you sign our data privacy agreement, or add the protections of a standard such as the AI Safety and Privacy Standard for Schools?

Federal rules that still apply

  • FERPA. When a vendor receives education records under the school official exception, the school must have direct control over how the vendor uses them. The U.S. Department of Education's model terms of service guidance is a practical checklist.
  • COPPA. The FTC's amended COPPA Rule, published on 22 April 2025, required most regulated companies to comply by 22 April 2026. It matters for any tool used by children under 13.
  • Federal AI guidance. The Department of Education's July 2025 Dear Colleague letter on using grant funds for AI lists protecting student data among its principles.

This is general information, not legal advice. Check state law and your district counsel's guidance before signing. Our COPPA compliance checklist and FERPA-aligned communication guide go into more detail.

Next steps for district and school leaders

  1. List every AI tool in use, including free teacher accounts, and who approved each one.
  2. For tools covered by agreements such as this one, confirm whether the protections apply to your contract and how to request them.
  3. Send the vendor questions above to every other AI provider and record the answers.
  4. Update your AI use policy so staff know which tools are approved and that humans make decisions about students. Our school AI use policy checklist can help.
  5. Train staff on what data they may enter into AI tools.
  6. Review the list each year, or when a vendor changes its terms.

How Borderset approaches student data

Borderset is a school management system, so we hold the student records these questions are about. Role-based access and audit trails control who can see student profiles, guardians, attendance and exam results, and student tracking keeps consent records with each student. Our guide to role-based access explains the principles. If you would like to review how Borderset handles data for your school, book a demo.

For writing school rules on AI, read the school AI use policy checklist. For where AI helps in operations, see AI in K-12 school operations. For other vendor changes, see Google's K-12 AI opt-in changes, and for privacy basics, the COPPA compliance checklist for K-12 schools.

Frequently asked questions

What did Anthropic agree to for schools?

On 10 November 2026, Anthropic, the AFT and the UFT announced that Anthropic adopted the AI Safety and Privacy Standard for Schools for districts using Claude for Teachers, making its data, oversight and transparency commitments contractually enforceable.

Does Anthropic use student or teacher data to train AI under the agreement?

No. The standard bars using student, educator or school data, including de-identified data, to train or improve models, with a narrow and auditable exception for safety and security purposes.

What is Claude for Teachers?

A version of Claude for verified U.S. K-12 educators, launched in July 2026, offering free access, teaching skills and connections to standards-aligned curricula. It does not include student-facing tools.

Does the agreement apply automatically to my district?

The AFT says it applies to Anthropic's existing agreements with districts using Claude for Teachers. Education Week reports districts must opt in for future licensing agreements, so confirm with the vendor.

What questions should schools ask AI vendors?

Ask whether your data trains models, who owns and can delete it, where it is stored, what safety controls exist, whether educators keep decisions, which certifications the vendor holds and how fast it reports breaches.

Does this agreement replace FERPA or COPPA?

No. The standard operates alongside FERPA, COPPA, IDEA and state privacy laws, and those laws still apply in full.

Back to all posts