Anthropic Signs Teachers Union Pact: 180-Day Data Deletion
Anthropic has signed the AI Safety and Privacy Standard for Schools. This is a legally binding agreement created with the American Federation of Teachers (AFT) and the United Federation of Teachers (UFT). The two unions announced the agreement on 10 November. The 29-page agreement says that when a school asks Anthropic to delete its data, the company must permanently remove it from active systems within 180 days. It must also remove all backup copies within 180 days.
The agreement also says Anthropic cannot use student or teacher data to train or improve any AI model at any time or for any reason, except for one limited safety-related purpose.
K-12 Dive reported the agreement on 10 November. The AFT says it has 1.875 million members, while the UFT has 200,000. These numbers show union membership, not how many people the agreement protects. The unions say the agreement applies to thousands of U.S. K-12 schools through existing agreements between Anthropic and school districts. Neither the AFT announcement nor the agreement mentions any payment.
The key insight: The clearest promise in the signed agreement is the 180-day data deletion deadline. When a school makes a request, Anthropic must remove its data from active systems and backups within 180 days. The company also cannot use that data to train AI models, except for a limited safety-related purpose.
What Anthropic Signed
The agreement is a formal document between Anthropic, PBC and the National Academy for AI Instruction. This academy is based in New York City and trains teachers. Its founding partners are the AFT, the UFT, Anthropic, Microsoft and OpenAI. The agreement lists ten principles. The first bans the use of student and teacher data to train AI models. The final principle is called 'Long-Term Data Responsibility'. The agreement says that following some principles but ignoring others does not count as compliance.
The ban on training covers everything users provide or produce, including prompts, answers, uploaded files, behavior signals and metadata. It also covers data that has been stripped of personal details or combined into larger groups. The agreement clearly states that removing identifying information does not create an exception. The ban also applies to data Anthropic collected before the agreement. It will remain in place even after the agreement ends, with no end date.
There is one exception. Anthropic can use student data for safety and security work only. Examples include finding signs of self-harm, grooming or security problems. The company cannot use this data to improve a general-purpose model. If the academy asks, Anthropic must provide a yearly statement signed by a senior company officer. This statement must confirm that the company has not used protected data for training outside the safety exception.
The Deadlines in the Agreement
The agreement sets deadlines for most of its promises. If Anthropic discovers a confirmed or suspected data breach, it must tell the affected school within 72 hours of becoming aware of it. Any school district that asks for the agreement's protections within 90 days of the effective date must be offered them. These protections can be added to the district's existing contract.
If the agreement ends, schools must receive at least 30 days to download and save their data for free before Anthropic deletes it. The agreement lasts two years and must be renewed in writing. Anthropic can leave by giving 60 days' written notice. The academy can also end Anthropic's participation if the company seriously breaks the agreement and does not fix the problem.
The agreement also lists independent security and privacy certifications that Anthropic must have or be working toward. These include SOC 2 Type II, ISO 27001, ISO 27701 and ISO 42001. It also requires FedRAMP Moderate when a customer asks for it.
Who the Agreement Covers
An additional document limits the agreement to one product for now. It says Anthropic's promises currently apply to Claude for Teachers. This service is not available to students or anyone under 18. As a result, rules that apply only to student users do not apply to this product. However, the document also says that all protections for covered data still apply, including student data that teachers enter into the service.
If Anthropic later offers a product to K-12 students in the United States, that product must follow the standard before it becomes available. The same additional document allows Anthropic to use combined results from its Anthropic Insights tool to carry out and publish research and analysis about how AI models are used, including in education.
Anthropic launched Claude for Teachers in July. According to the AFT announcement, verified K-12 teachers in the United States can use Claude for free. The announcement also says the new agreement automatically applies to Anthropic's existing agreements with school districts that use Claude for Teachers.
Where the Unions' Efforts Stand
Microsoft signed a similar agreement with the two unions in September. It took effect on 1 November, according to K-12 Dive. The unions say Anthropic had already agreed to follow the standard. They say 10 November is the date when the agreement officially takes effect.
AFT President Randi Weingarten said the agreement brings them one step closer to making this an industry standard, now that Anthropic has joined them. Elizabeth Kelly, Anthropic's head of beneficial deployments, 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.'
Companies charge different prices for AI services in schools. Google lists its Google AI Pro for Education add-on at $20.00 per user each month, as our 4 November article reported. By comparison, verified teachers can use Claude for Teachers for free. The agreement comes as Anthropic's IPO filing is being watched. Our article about its S-1/A filing explains this further.
What the Agreement Means in Practice
In our reading, the agreement covers less than the headline might suggest. The additional document limits Anthropic's promises to Claude for Teachers. This is a product for adult teachers. It also says that rules applying only to student users do not apply.
In our view, the agreement turns the unions' safety standard into a set of contract rules with clear deadlines. Anthropic has 72 hours to report a data breach, 90 days to offer the terms to school districts and 180 days to delete data after a request.
The signed agreement, Principle 1
'at any time, for any purpose'
Three Important Effects
FOR SCHOOL DISTRICTS School districts can ask to add the agreement's protections to their existing contracts. The document says Anthropic must make these protections available within 90 days of the effective date.
FOR ANTHROPIC The company's promises currently cover Claude for Teachers. The additional document says that any product Anthropic later offers to K-12 students in the United States must meet the standard before it is released.
FOR OTHER AI COMPANIES Microsoft and Anthropic have now signed the standard. OpenAI is also a founding partner of the academy, but the announcement does not list it as a company that has signed the agreement.
The Business Engineer View
This story connects to the Business Engineer framework called Inside Anthropic's Permission Layer.
The framework explains the role of the governance layer this way: 'It is where decisions are made about who gets access to frontier intelligence, under what conditions, and for which purposes.'
In our view, the school agreement is an example of this layer written into a contract. It explains who can use student data, what they can use it for and how long they can keep it. The framework was created before this agreement and does not discuss schools or the AFT.
What We Still Do Not Know
The announcement and agreement do not say how many school districts or teachers use Claude for Teachers. This means the total number of people protected by the agreement is not publicly known. Neither document says whether OpenAI, the third AI company among the academy's founding partners, will sign the same standard.
The two parties must complete two additional documents within 90 days of the effective date. One covers the academy's setup, and the other covers technical requirements. In the published version, the spaces for initials and the date on the first document are still blank.
The Bottom Line
Anthropic has signed an agreement with the AFT and UFT promising to delete school data within 180 days of a request. It must also report data breaches within 72 hours and cannot use student or teacher data to train AI models, except for a limited safety-related purpose. For now, the agreement covers Claude for Teachers, a service for adult educators. Neither the announcement nor the agreement mentions any payment.
A note about sources. The agreement's rules come from the signed memorandum published by the AFT with its announcement on 10 November. Membership numbers and quotes come from that announcement. K-12 Dive first reported the agreement. The conversions from hours to days are our own calculations.
Sources: AFT press release, 'AFT, UFT and Anthropic Sign Binding AI Safety and Privacy Agreement for Schools', 10 November 2026 · AI Safety & Privacy Standard for Schools, memorandum of agreement signed by Anthropic, PBC (PDF, published by the AFT) · K-12 Dive, 'Anthropic enters student AI privacy agreement with teachers unions', 10 November 2026 (via Yahoo News)