Schools are no longer asking whether artificial intelligence will enter classrooms. They are deciding which tools get in, which grades are too young, what counts as a learning use, and how much student data a vendor should ever touch. For parents, that changes the useful question. It is less "Is AI good or bad for kids?" and more "What exactly is my child being asked to use, at what age, under whose supervision, and with what data protections?"

A classroom desk with a laptop, parent checklist notebook and privacy shield symbol representing questions about school AI tools for children.

That distinction matters because the latest school policies are not all moving in the same direction. On September 9, 2026, Microsoft, the American Federation of Teachers and the United Federation of Teachers announced a National AI Safety & Privacy Standard for schools. The agreement says school AI contracts should bar companies from using student and educator data to train AI models, require plain-language transparency for families and educators, prevent student tracking, keep humans in charge of high-stakes decisions, and give districts enforcement tools if vendors break the rules.

At almost the same time, large districts are pulling back from student-facing AI, especially for younger children. New York City Public Schools says that for the 2026-27 school year, student-facing generative AI is banned in grades 2K through 8, while high school use is limited to approved, supervised pilots and AI literacy lessons. Los Angeles Unified has also moved to restrict generative AI on district-owned devices, while still allowing some non-generative tools, such as speech recognition, text-to-speech, captioning, translation and adaptive learning systems.

Parents do not need to become software auditors. But they do need a sharper checklist than "the school approved it." School approval can be meaningful, especially when a district has a serious privacy review. It can also hide practical details that matter at home: whether the tool talks directly to a child, whether it stores voice or writing samples, whether a teacher reviews the output, whether the child can use it to bypass thinking, and whether the product is optional for students with different needs or family concerns.

The current shift is about student-facing AI, not every classroom technology

A lot of confusion comes from using the word AI for very different things. A teacher using AI after school to draft a worksheet is not the same as a seven-year-old chatting with a generative tutor. A screen reader, live captioning tool or translation aid is not the same as a companion chatbot. A math platform that adjusts problem difficulty is not the same as a tool that writes a paragraph for the student.

For family decision-making, the most important category is student-facing generative AI. That means a child directly prompts a system that can produce new text, images, audio, explanations or feedback. It may look like a tutor, writing coach, reading assistant, brainstorming partner or classroom chatbot. These tools can be useful, but they also introduce risks that ordinary educational software did not create in the same way. They may generate errors confidently. They may give too much help. They may invite personal disclosure. They may collect rich samples of a child's thinking, voice, writing style, questions and mistakes.

New York City's policy draws that line plainly. It restricts student-facing generative AI in grades 2K-8 while leaving room for assistive technology, legally required supports under IEPs and 504 plans, teacher-led classroom technology, approved assessments, e-books, coding, robotics and other specific exceptions. That is a useful model for parents even outside New York: ask whether the policy limits the AI interaction itself or simply says "technology" in a broad way. Broad language can sound strict while leaving the most important details unclear.

Age changes the trade-off

A high school junior using a vetted tool during a supervised career and technical education project is in a different situation from a second grader using a chatbot during reading time. Age does not make every risk disappear, but it changes what a child can understand and what supervision has to do.

For preschool through early elementary children, the case for direct generative AI is weak unless the use is tied to an accessibility need or a very narrow, teacher-controlled activity. Young children are still building the habit of asking people for help, listening to real voices, noticing facial expressions, handling boredom, sounding out words, using their hands, and struggling through a problem long enough to learn from it. A chatbot that is always available, always patient and always able to supply an answer can undercut those ordinary learning moments if it becomes a routine substitute.

For upper elementary and middle school students, the question becomes whether the AI is giving feedback that keeps the student doing the thinking. A grammar nudge, vocabulary support or step-by-step prompt can help. A tool that drafts the answer, summarizes the reading before the child has tried, or turns every assignment into prompt management may be too much assistance. This is also the age when children may treat a conversational system as more socially meaningful than adults intend. Parents should be especially cautious about open-ended chatbots, companion-style interfaces and tools that encourage personal confession.

For high school students, AI literacy is no longer optional. They will meet these tools in college, work, search, creative software and ordinary productivity apps. A blanket "never use it" rule is hard to maintain and may leave teens less prepared to judge AI outputs. But the school version should still be bounded. Students should know when AI is allowed, how they must cite or disclose its use, what data they should never enter, and when the assignment is meant to test unaided reading, writing, calculation or reasoning.

A practical parent question is: "Is my child learning about AI, learning with AI, or being taught by AI?" Those are three different uses. Younger children can learn basic AI concepts without using a chatbot. Older students can use AI as a tool while still being assessed on their own thinking. Very few children need AI to become the main instructor.

Privacy is not just a policy link at the bottom of a login page

Student privacy has a legal frame in the United States, but the parent-facing reality is often messy. COPPA generally requires verifiable parental consent before online services collect personal information from children under 13, with exceptions. The Federal Trade Commission explains that schools may consent on behalf of parents when a service is used for the school's benefit and no other commercial purpose. But the FTC also says the operator must give the school direct notice of what it collects, how it uses and discloses the data, and must allow review, deletion and limits on further collection.

That means a school can sometimes approve an edtech tool without sending every parent a separate consent form. It does not mean a vendor can use children's data for advertising, build unrelated commercial profiles, keep data forever, or treat school consent as a blank check. It also does not mean parents are wrong to ask for a plain-language explanation.

AI raises the stakes because the inputs can be unusually revealing. A worksheet app might store scores and answers. A generative tool may collect drafts, voice recordings, prompts, emotional questions, reading struggles, learning accommodations, behavioral patterns or uploaded files. Even when a company promises not to train its main AI model on student data, parents should still ask what logs are kept, who can review them, how long they remain, whether subcontractors process them, and whether the system uses the data to personalize future responses.

The Microsoft-AFT-UFT standard is notable because it tries to turn broad privacy principles into contract terms. Its announced protections include no training on student or educator data, no sale or repurposing of that data, limits on collection and retention, transparency about product changes, security requirements, and human oversight for decisions that affect students. Parents can use that standard as a benchmark even if their school uses another vendor. If a company cannot answer those questions clearly, the issue is not that parents are being difficult. The tool is not ready for quiet adoption.

The five questions worth asking before a classroom AI tool becomes routine

Parents usually do not get to negotiate a district contract, but they can ask questions that force the practical details into view. Start with these.

  1. What does the child actually do with the tool?

Ask for the student workflow, not just the product name. Does the child type free-form prompts? Speak into a microphone? Upload writing? Receive automated feedback? Chat with a character? Use it once a month, once a week or every day? Is it teacher-led on a shared screen or one-to-one on a device?

This question separates a narrow classroom aid from a general chatbot. It also helps parents understand whether the tool changes the assignment. A reading assistant that listens to fluency for five minutes is not the same as a bot that explains every paragraph of a book before the child has tried to make sense of it.

  1. What data is collected, and what is forbidden?

The useful answer names categories: account identifiers, grade, school, typed prompts, uploaded documents, audio, handwriting, performance data, usage logs, device data and teacher feedback. It should also say what students are told never to enter, such as home address, phone number, private family information, medical details, passwords, photos of other students or anything they would not want stored in a school record.

If the school says "no personal data," ask what that means. A child's essay about a family event can contain personal data. A voice sample can be personal data. A learning accommodation can be sensitive. The privacy conversation should match the actual classroom use.

  1. Who sees the output, and who is accountable for mistakes?

Generative AI can be wrong, biased, overconfident or inappropriate. A school tool should have an adult accountability chain. Does the teacher review AI feedback before it reaches the student, or does the system respond directly? Can students flag a bad answer? Are logs reviewed? Is there a process for correcting records or removing harmful content?

This matters most when the tool affects grading, placement, discipline, special education planning or intervention decisions. Families should be wary of any system that turns AI output into a decision about a child without meaningful human review. The strongest new school AI standards make human oversight central for high-stakes uses.

  1. What is the learning purpose, and how will the school know it worked?

"Personalized learning" is not enough. Ask what specific problem the tool is meant to solve. Is it helping English learners practice vocabulary? Giving math hints during independent work? Supporting students who need text-to-speech? Teaching older students how to critique AI output? Reducing teacher paperwork?

Those are different goals. A tool that saves teachers planning time may be valuable, but it should not be described as a direct student learning benefit unless the school can explain the student-facing evidence. A tool that helps a small group of students access material may be worthwhile even if it is not needed by the whole class. The more intrusive the tool, the clearer the benefit should be.

  1. Can families opt out, and what happens next?

An opt-out that leaves a child with busywork is not a real alternative. Ask what assignment, support or device-free path is available if a family declines a student-facing AI tool. Also ask whether opting out affects grades, classroom participation, access to advanced coursework or required services.

There will be cases where a tool is part of a required accommodation or assessment system, and the answer may be more complicated. But for general classroom AI, schools should be able to explain the alternative without making the child feel singled out.

Parents often fold classroom AI into the larger screen-time argument. That is understandable. If a child already spends hours on school devices, a chatbot may feel like one more digital layer. But AI policy and screen policy should not be treated as the same thing.

New York City's 2026-27 guidance shows the distinction. It sets a student-facing generative AI moratorium for grades 2K-8, while separately recommending no one-to-one screen time for grades 2K-2, no more than 30 minutes of daily one-to-one screen time for grades 3-5, and no more than 45 minutes for grades 6-8, with exceptions. High school screen use is course dependent. Assistive technology remains available when required.

That structure is more useful than a single household-style timer because school screens vary. A child using a communication device, a dyslexic student using text-to-speech, a class using a coding simulation and a student chatting with a generative bot are not having the same experience. Parents can support limits on unnecessary screens while still defending necessary access tools.

At home, the same distinction helps. If a school introduces AI, families may need a short house rule: school AI stays in school accounts; no personal secrets or family information go into prompts; AI can help explain a concept but not complete the assignment; and any confusing or upsetting response gets shown to an adult. That is more practical than trying to monitor every feature inside every app.

Teacher use can be helpful, but it still needs boundaries

Most parent concern focuses on children using AI directly. Teacher use deserves a calmer but still serious look. A 2026 NPR/Ipsos poll of K-12 teachers found that many teachers who use AI say it makes them more productive, especially for creating classroom materials and planning lessons. The same poll found broad concern about AI's effect on students' critical thinking, and only about a third of teachers reported formal school guidelines for student AI use.

Those findings point to a sensible middle ground. Teacher-facing AI can reduce repetitive administrative work and help produce drafts of materials, quizzes, rubrics or parent communications. But schools should be careful about using AI to grade student writing, detect cheating, evaluate behavior, write special education documents, or generate sensitive messages without close human review.

Parents can ask whether teachers are allowed to put student work into public AI systems. A teacher pasting a child's essay, IEP notes or behavior report into an unapproved tool may create a privacy problem even if students never touch AI themselves. Districts should have a simple approved-tools list and a rule against entering identifiable student data into tools that have not passed review.

What a decent school notice should include

A useful notice to families does not need to be long. It should name the tool, the vendor, the grades involved, the learning purpose, the frequency of use, the data collected, the data retention period, whether the vendor uses data for model training, whether the tool includes open-ended chat, whether student interactions are logged, who can access those logs, what human review exists, and whether families can opt out.

It should also say what the tool will not do. For example: it will not be used for grading final essays; it will not make placement decisions; it will not replace required services; it will not be used as a companion chatbot; it will not collect location data; it will not sell or repurpose student data; it will not train commercial models on student inputs.

The wording matters. Families should not have to decode a vendor privacy policy written for lawyers. One of the strongest parts of the new school AI privacy standard is the insistence on plain-language answers for educators and parents. If a school cannot explain a tool in ordinary language, it probably has not finished deciding how the tool belongs in a classroom.

When to push back harder

Not every imperfect notice requires a campaign. Schools are learning, teachers are under pressure, and some tools will be adjusted as policies mature. Still, several warning signs deserve firmer questions.

Push back if a tool is introduced to young children as a friendly companion rather than a limited learning aid. Push back if the vendor says student data may be used to train models or improve unrelated products. Push back if the product requires open-ended chat for grades where the school cannot explain why it is developmentally appropriate. Push back if students are asked to enter personal reflections, mental health concerns or family details into an AI system. Push back if AI output affects grades, discipline, placement or services without clear human review.

Also push back if the school treats opt-out requests as hostile. A parent asking for the privacy notice, the approved-tools list or the alternative assignment is not rejecting teachers. They are doing ordinary due diligence for a child. Schools that want trust should make those answers easy to find.

What to say at home without turning AI into a forbidden mystery

Children handle technology rules better when the adult explanation is concrete. For younger kids, the message can be simple: "Some computer tools can answer like a person, but they are not people. They can make mistakes. If a school tool asks you something private or makes you feel strange, tell me or your teacher."

For middle schoolers, add the academic boundary: "Use help to understand, not to pretend. If the tool gives you the answer, you still need to be able to explain it in your own words. Do not put private information about yourself, friends or family into a school app unless a teacher has clearly explained why."

For high schoolers, treat AI as a literacy issue. Ask them to show you the tools they are allowed to use. Talk through one example of a bad AI answer. Discuss disclosure: when a teacher expects original work, when brainstorming is allowed, and when AI use must be cited. Teenagers do not need panic. They need standards they can actually apply when a shortcut is one tab away.

The practical bottom line

The best school AI policies are not anti-technology. They are pro-clarity. They separate teacher tools from student tools, accessibility supports from convenience features, high school literacy from early-childhood experimentation, and supervised learning aids from open-ended companions. They also treat student data as something borrowed for a narrow educational purpose, not as raw material for product development.

For parents, the goal is not to win an abstract argument about AI. It is to make sure a child is not quietly enrolled in an experiment that nobody can explain. Ask what the tool does. Ask what it collects. Ask who checks it. Ask what learning problem it solves. Ask what happens if your family says no.

A school that has good answers will be able to give them plainly. A school that does not have those answers may need more time before putting a chatbot in front of children.