Anthropic Signs Binding School Data Pact With Teachers Unions

Human Verified 0

Share to: , and

A teacher's desk with a laptop in a sunlit classroom

Anthropic has signed a legally binding agreement with the American Federation of Teachers and its New York City affiliate that makes AI safety, privacy and transparency standards enforceable in thousands of US schools. The pact, announced on Tuesday, applies the unions’ national AI Safety and Privacy Standard for Schools to Anthropic’s existing deals with districts using Claude for Teachers. It bars the company from using student or educator data to train its models, gives schools control over how long data is kept, and requires plain-language disclosure of what its tools collect.

What the agreement requires

The agreement rests on three principles the unions say are now enforceable. Student data is off-limits: information from students and educators may be used only for authorised educational purposes, never sold and never used for targeted advertising. Schools decide how long the data is kept and when it is deleted.

Educators stay in charge, with safeguards built for schools and teachers, not AI systems, responsible for instructional decisions about students. And there are no black boxes: schools, educators and families must get clear, plain-language information about how the tools work, what data they collect, where they fall short and what safeguards apply.

Behind the announcement sits a 29-page memorandum signed by Anthropic and published by the unions, setting out ten principles that run for two years and must be renewed in writing to continue. The document is the standard itself, written by the unions’ National Academy for AI Instruction.

When a school asks for its data to be deleted, Anthropic must permanently remove it from active systems within 180 days, with backup copies deleted or made commercially unrecoverable on the same deadline. A confirmed or suspected breach must be reported to the affected school within 72 hours.

The training ban, and its one exception

The first principle is the sharpest: no student or educator data may be used to train, fine-tune, update or benchmark any model, at any time, for any purpose. The ban covers prompts, responses, uploaded files and metadata, and applies to de-identified data too.

It reaches back to data collected before the agreement, and it survives if the agreement ends. The only exception is a narrow safety one, for detecting or stopping self-harm, abuse, grooming, threats or attacks, using only the minimum necessary. Each year a senior officer must confirm in writing that no covered data crossed from safety systems into any other model.

Second to sign, with others still outside

Anthropic is the second major AI company to sign the standard. Microsoft agreed to the same terms in September, and its agreement took effect on 1 November, letting any US district add the protections to a new or existing Microsoft contract without renegotiating.

OpenAI is still in talks with the unions, while Google and Meta have no agreement at all. Anthropic, Microsoft and OpenAI are founding partners of the National Academy for AI Instruction, the training centre the unions set up with them in New York.

The stakes are visible in the districts that moved first. New York City imposed a one-year moratorium on student-facing AI for children from pre-kindergarten through eighth grade in September, covering nearly 600,000 students, and Los Angeles restricted access for the 2026-27 school year. Los Angeles officials had hoped pupils could keep using Search with AI switched off.

A school hallway with lockers and a backpack on a bench

Why It Matters

The deal matters because it turns a voluntary promise into a contract schools can enforce, in a space where federal and state rules have barely arrived. The unions’ president said the standard should be written into law — but until it is, the agreement gives districts a remedy the law does not.

A breach can end a vendor’s place in the programme, and the terms, including the training ban, sit in the district’s contract rather than a company’s policy page. That is a different kind of accountability from the courtroom route: nonprofits sued OpenAI, Google and Anthropic to halt frontier AI, while the unions chose to bind the companies with paper instead.

The timing also sharpens the contrast between the labs. Anthropic signs days before its shares are due to be priced on 18 November, accepting binding limits on how it can use school data just as it asks the market to value it at more than a trillion dollars.

OpenAI, still negotiating, and Google and Meta, outside the standard entirely, now face the question the unions are asking in public: why sign later, or not at all, when the terms are already written? The answer will show whether the standard becomes an industry norm or stays a badge worn by two companies.

One limit keeps the deal narrower than its headline suggests. For now the terms apply only to Claude for Teachers, a tool built for adult educators rather than students, so the rules on students’ direct use of AI do not bite yet.

Student information that teachers type into the tool is still protected, but the standard will only reach pupils directly if Anthropic extends it to a student-facing product. The industry’s other training-data promise raises the same question: Google, OpenAI, Microsoft and Anthropic have agreed to honour one robots.txt signal against AI training, yet the dates for doing so stretch into 2027.

Background

The unions have been building toward this since May, when the AFT’s ten-point plan called for a ban on student-facing AI in elementary school, a screen ban for kindergarten through second grade, and a total ban on social companion chatbots before sixteen. Anthropic launched Claude for Teachers in July, giving verified US educators free access.

Source: AFT announcement and K-12 Dive report

Share to: , and

Faheem R. Khattak

Senior automation engineer and AI marketing journo covering LLMs, algorithms, analytics, core updates, Bing search, and technical SEO.

0 Comments

Leave a Reply

Optional Take Tag:

🛡️ Spam protection: Honeypot + Time-Trap Active

This site uses Akismet to reduce spam. Learn how your comment data is processed.