Anthropic has signed the AI Safety and Privacy Standard for Schools, a binding agreement drafted with the American Federation of Teachers (AFT) and the United Federation of Teachers (UFT), the two unions announced on 10 November. The 29-page agreement says that once a school asks for its data to be deleted, Anthropic must “permanently delete it from active systems within 180 days”, with backups purged within 180 days as well.
It also bars the use of student or educator data to train or improve any AI model “at any time, for any purpose”, apart from one narrow safety exception.
K-12 Dive reported the deal on 10 November. The AFT release puts the AFT’s membership at 1.875 million and the UFT’s at 200,000. Those are union membership figures, not a count of people the agreement covers: the unions say it applies to “thousands of U.S. K-12 schools” through Anthropic’s existing district agreements. Neither the AFT release nor the agreement text names any payment.
Business Pill · Guardrails Explained
An evergreen Business Engineer explainer on guardrails. It explains the general idea only and does not discuss Anthropic’s school agreement.
The key insight: As we read it, the 180-day deletion clock is the most concrete promise in the signed text: once a school asks, Anthropic must remove its data from active systems and backups within 180 days, and it may never train models on that data outside a narrow safety exception.
The agreement is a memorandum between Anthropic, PBC and the National Academy for AI Instruction, a New York City-based teacher-training academy whose founding partners are the AFT, the UFT, Anthropic, Microsoft and OpenAI. It sets ten principles, beginning with a ban on training AI models on student or educator data and ending with ‘Long-Term Data Responsibility’, and says complying with some but not others “is not compliance”.
The training ban covers prompts, responses, uploaded files, behavioral signals and metadata. It also covers de-identified and aggregated data: “De-identification is not a loophole”, the text says. The ban applies retroactively to data already collected and survives the end of the agreement indefinitely.
The exception allows student data to be used only for safety and security work, such as detecting self-harm risk, grooming or security incidents, and not to improve a general-purpose model. On request, Anthropic must give the academy a yearly statement, signed by a senior officer, confirming that no covered data was used for training outside that exception.
Deadlines in the AI Safety & Privacy Standard for Schools signed by Anthropic, in days: deletion from active systems and backups within 180 days of a school’s request; protections offered to districts within 90 days of the effective date; 60 days’ written notice for Anthropic to leave; at least 30 days to export data after termination; breach notice within 72 hours (3 days, our arithmetic). Source: the signed memorandum of agreement published by the AFT, 10 November 2026.
The Deadlines in the Text
The agreement attaches a clock to most of its promises. Anthropic must tell a school about a confirmed or suspected breach no later than 72 hours after becoming aware of it. It must offer the agreement’s protections to any district that asks within 90 days of the effective date, and those terms can be added to a district’s existing contract.
If the arrangement ends, schools get at least 30 days to export their data at no charge before it is deleted. The term is two years and must be renewed in writing. Anthropic can leave with “60 days’ written notice”, and the academy can end Anthropic’s participation for a material breach that is not cured.
The text also lists the independent certifications Anthropic must hold or be pursuing, including SOC 2 Type II, ISO 27001, ISO 27701 and ISO 42001, with FedRAMP Moderate where a customer requires it.
Who It Covers
An addendum limits the scope to one product for now. It says Anthropic’s commitments “currently apply to Claude for Teachers”, which “is not offered to students or individuals under 18”, so rules that apply only to student users do not apply. “All protections for Covered Data, including student data entered by educators, apply in full”, it adds.
If Anthropic later offers a product to U.S. K-12 students, the addendum says it must meet the standard before it is offered. The same addendum lets Anthropic use aggregated outputs from its Anthropic Insights tool for “conducting and publishing research and analysis regarding AI model use”, including use in education.
Anthropic launched Claude for Teachers in July. The AFT release says it gives verified U.S. K-12 educators free access to Claude, and that the agreement “automatically applies to Anthropic’s existing agreements with school districts using Claude for Teachers”.
Where the Unions’ Push Stands
Microsoft signed a similar agreement with the two unions in September, and it took effect on 1 November, K-12 Dive reported. The unions say Anthropic had already committed to the standard, and that 10 November marks the point it goes formally into effect.
AFT President Randi Weingarten said the agreement is “one step closer to an industry standard with Anthropic joining us”. Elizabeth Kelly, head of beneficial deployments at Anthropic, said: “That is only possible if teachers, students and school systems trust us.” UFT President Michael Mulgrew said: “Ideally, the federal government would step in, but since it hasn’t, teachers unions are working to hold AI companies accountable.”
The school market is being priced in different ways. Google lists a Google AI Pro for Education add-on at $20.00 per user a month, as our 4 November piece reported, while Claude for Teachers is free to verified educators. The agreement lands as Anthropic’s IPO filing is in view; our piece on its S-1/A covers that.
The Structural Read
As we read it, the scope is narrower than the headline deal: the addendum limits Anthropic’s commitments to Claude for Teachers, a product for adult educators, and says rules that apply only to student users do not apply.
As we read it, the agreement turns a union standard into contract terms with clocks attached: 72 hours for breach notice, 90 days to offer the terms to districts, 180 days to delete.
The signed agreement, Principle 1
“at any time, for any purpose”
Three Implications
FOR SCHOOL DISTRICTS Districts can ask for the agreement’s protections to be added to their existing contracts, and Anthropic must make them available within 90 days of the effective date, the text says.
FOR ANTHROPIC Its commitments currently cover Claude for Teachers; the addendum says any product later offered to U.S. K-12 students must meet the standard before it is offered.
FOR OTHER AI VENDORSMicrosoft and Anthropic have now signed; OpenAI, also a founding partner of the academy, is not named as a signatory in the release.
The framework says of the governance layer: “It is where decisions are made about who gets access to frontier intelligence, under what conditions, and for which purposes.”
As we read it, a school agreement is that layer written as a contract: it sets who may use student data, for which purposes and for how long. The framework predates this agreement and does not discuss schools or the AFT.
What Is Not Established
The release and the agreement do not say how many districts or teachers use Claude for Teachers, so the number of people covered is not public. Whether OpenAI, the third AI lab among the academy’s founding partners, will sign the same standard is not stated in either document.
Two addenda, one on the academy’s configuration and one on technical requirements, are to be completed by both parties within 90 days of the effective date; in the published copy, the initials and date lines of the first are blank.
This Week In The AI Supercycle
The Great Convergence of Capital, Power, and Geopolitics
This week’s issue: for two years the AI build-out was priced as a supply story, and last week as a credit story. This week it was priced as a permission story, and the answer arrived as a cost, not a ban.
Anthropic’s signed agreement with the AFT and UFT commits it to delete a school’s data within 180 days of a request, to give breach notice within 72 hours and never to train on student or educator data outside a narrow safety exception. For now it covers Claude for Teachers, a tool for adult educators, and no money is named in the release or the text.
A note on sourcing. The agreement terms are from the signed memorandum of agreement that the AFT published with its 10 November release; membership figures and quotes are from that release. K-12 Dive reported the agreement first. Conversions of hours to days are our arithmetic.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
