Status as of Nov. 14, 2026. We will update this guide here as more companies sign.

Saying there are no meaningful federal or state rules for AI in schools, the American Federation of Teachers and its New York City affiliate, the United Federation of Teachers, negotiated their own with AI companies. The AI Safety and Privacy Standard for Schools is a contract, not a pledge: districts that add it to their agreements can hold the vendor to it.

"We can get angrier and angrier, or we can act decisively; anything less than legally enforceable provisions is simply a wish list," AFT President Randi Weingarten said when Microsoft signed in September (Microsoft).

Who has signed

Company

Product covered

Status

How districts get the protections

Microsoft

Microsoft's education products

Signed Sept. 9, 2026; offered to every U.S. district since Nov. 1, 2026

Add to new or existing Microsoft agreements, with no renegotiation or renewal required

Anthropic

Claude for Teachers

Signed; in effect Nov. 10, 2026

Applies automatically to existing district agreements; districts opt in for future licenses

OpenAI

ChatGPT for Teachers

Still negotiating with the union

Not available yet; OpenAI offers its own 16-state privacy agreement

Google

Gemini for Education

No agreement with the AFT

Google's own Workspace for Education terms apply

Sources: Microsoft's announcement and fact sheet; the AFT; Education Week.

Microsoft: the first signer, and the broadest

Microsoft, the AFT and the UFT announced the standard on Sept. 9 after what Microsoft called "months of intense negotiations." Since Nov. 1, Microsoft has offered the protections to every school district in the United States. Districts can add them to new or existing agreements "with no contract renegotiation or renewal required," and the protections are "contractually enforceable if those commitments are not met," according to Microsoft's fact sheet.

"This standard sets a high bar for child privacy and AI safety, and we'll extend this agreement to every school district across the country," said Brad Smith, Microsoft's vice chair and president (Microsoft). Districts can request to add the standard to their Microsoft agreement through a contact form linked from the announcement.

Anthropic: narrower, because the product is for teachers only

The AFT and UFT announced Anthropic's agreement on Nov. 10. It applies automatically to Anthropic's existing agreements with districts using Claude for Teachers, the free tool for verified U.S. K-12 educators that Anthropic launched in July (AFT).

Education Week notes three limits. The agreement is narrower than Microsoft's because Claude for Teachers has no student-facing tools. It covers only Claude for Teachers, not education tools other companies build on Claude's models. And districts must opt in for future licensing agreements (Education Week).

"That is only possible if teachers, students and school systems trust us," said Elizabeth Kelly, Anthropic's head of beneficial deployments. "These agreements provide a foundation for that trust" (AFT).

OpenAI: negotiating, with its own privacy agreement in the meantime

The union is still negotiating with OpenAI, Education Week reported on Nov. 10 (Education Week). OpenAI is a founding partner, with Anthropic and Microsoft, of the AFT's National Academy for AI Instruction (AFT).

In the meantime, OpenAI offers its own route. In August, it expanded ChatGPT for Teachers to 55 more school systems and announced a data privacy agreement covering 16 states, designed so districts don't have to negotiate separate privacy terms. It says information shared in ChatGPT for Teachers workspaces is not used to train its models by default, and that the product is free for verified U.S. K-12 educators through June 2028 (citybiz). Those are OpenAI's terms, not the union standard, so compare them point by point.

Google: no agreement

The AFT has no agreement with Google or Meta (Education Week). Google's own terms say chats and uploaded files from Gemini app users in Workspace for Education editions "won't be reviewed by human reviewers or otherwise used to improve generative AI models" (Google). That covers training, but it is not the same as the union standard's other nine protections or its enforcement terms. Google is also changing its K-12 AI defaults; see our guide to its new controls.

What the standard requires

Microsoft's fact sheet lists 10 contractually enforceable protections (Microsoft):

#

Protection

What it means

1

No training on student or educator data

Data can't train or improve an AI model, apart from a narrow safety and security exception

2

Data minimization by design

Products collect only what they need; tracking, profiling, biometrics and memory need limits and review

3

Customer data ownership

Data can't be sold or used for ads or product development; customers control export, retention and deletion

4

Human oversight and explainability

High-risk decisions require human review; AI companions are prohibited

5

Real accountability

Providers keep certifications, report breaches within 72 hours and fix violations, or AI use is paused or ended

6

Security that matches the stakes

Encryption, access controls, independent testing and incident-response plans, including for subprocessors

7

Consent and transparency for families

Plain-language guides on how the products work and on privacy, safety and security

8

Equity, accessibility and inclusion

Products must be accessible and fair; providers disclose and fix gaps and fund an independent study

9

No feature creep and no vendor lock-in

Customers control material new features and can export data and switch providers

10

Long-term data responsibility

Training and sale bans are permanent; deletion and breach duties continue after the contract ends

GovTech's reading of the agreement adds detail. It restricts behavioral tracking, long-term profiling, keystroke logging and continuous attention monitoring. Providers generally have 180 days to remove data from active systems after a district asks. AI systems can't independently make decisions on student discipline, academic placement or employee evaluation without meaningful human review. And districts can end agreements when violations aren't resolved (GovTech).

Other companies in the AFT's AI academy can adopt the standard, but they must agree to all of its principles and offer equivalent protections to education customers that ask (GovTech).

What it doesn't do

The standard is a contract between a union and individual companies, not a law. Weingarten has said the union wants it "embedded in law" (AFT). The protections apply only to the products named in each agreement and only once a district has them in its contract. It will still be mostly up to districts to secure AI protections (Education Week).

Sources

  1. AFT, UFT and Microsoft announce ‘National AI Safety & Privacy Standard’ for schools to protect students, families and educators. Microsoft (2026)
  2. AI Safety & Privacy Standard for Schools (fact sheet). Microsoft (2026)
  3. AFT, UFT and Anthropic Sign Binding AI Safety and Privacy Agreement for Schools. American Federation of Teachers (2026)
  4. Anthropic Agrees to Protections for Teachers Who Use Claude. Education Week (2026)
  5. Microsoft, AFT, UFT Set Precedent for District AI Governance. GovTech (2026)
  6. OpenAI Expands Free ChatGPT for Teachers Program to 55 More U.S. School Systems. citybiz (2026)
  7. Google Workspace with Gemini FAQ for Education. Google (2026)