Data Privacy Standards for AI Children's Education Apps

AI education apps put parents in a bind that no amount of careful reading can fully resolve. The same data that makes an app good at teaching, voice recordings of a child sounding out words, logs of which problems they get wrong, how fast they move through a lesson, whether they have a disability that shapes how they learn, is also the data that's most valuable to advertisers and most damaging if it leaks. It's a structural feature baked into what adaptive learning actually requires, not a flaw a company can patch.

None of this needs a villain to explain it. E|Research out of NYU (Nair and Greenstadt) found a consistent pattern across edtech organizations: privacy gets acknowledged as important, then gets deferred constantly, in favor of shipping features, growing the user base, and hitting funding milestones. Nobody sits down and decides to treat children's data carelessly. There's just no mechanism inside most of these companies that forces privacy to the front of the line before a product ships.

The responsibility also has a way of evaporating between the people who could catch a problem. Companies point to their cloud providers. Schools assume the vendor has already handled compliance. Parents assume the school already vetted the tool before assigning it. Each link in that chain is leaning on the one before it, and if one link is wrong, nobody downstream finds out until something breaks.

O|The vast majority of edtech tools share children's personal information with third parties, Internet Safety Labs found. Not a scandal at a handful of bad-actor companies: a baseline condition of the market itself. Any framework for evaluating these apps has to start from that fact, not treat it as the exception that a smart parent might get lucky and avoid.

D|What the law covers, and where parents lack protection

Federal law gives parents a real floor to stand on, just not a very high one. The law provides a real but incomplete floor: parents who understand what COPPA, FERPA, and the amended 2026 rules actually cover will see exactly where the gaps are.

FERPA covers education records held by schools and by the vendors acting on a school's behalf. It's the law that's supposed to keep a student's grades and disciplinary history from wandering off. It was written long before generative AI existed, and it says nothing specific about what happens when a school's reading app feeds a child's voice into a machine learning model.

COPPA is the one that actually targets apps built for children under 13, and it requires verifiable parental consent before an operator collects personal information from a young child. Here's where most of the confusion lives: schools can consent on a parent's behalf under what's known as the school authorization exception. That exception rests on FTC guidance rather than codified rule text, since the FTC chose not to formalize it in the 2025 COPPA amendments, and it only covers educational use. It does not authorize a company to train AI models on student data, build marketing profiles, or hand data to partners outside the educational purpose. Plenty of schools, and plenty of parents, assume it covers more than it does.

The amended COPPA rule, which became mandatory on April 22, 2026, closed part of that gap and left the rest standing. For the first time, federal rule text addresses AI directly: using a child's personal information to train an AI system is "generally not considered integral to a website or online service," so an operator now needs separate consent just for that use. The amendment also named voiceprints, biometric identifiers pulled from a recording of a child's voice, as personal information in their own right, though it stops short of covering voice data more broadly.

States are filling in some of what federal law leaves out, unevenly. California, Colorado, and Connecticut have gone furthest with age-appropriate design codes, which regulate how a product is built rather than just what data it's allowed to collect. But coverage varies enormously by state, and a family in one part of the country has meaningfully different protections than a family in another.

None of this adds up to real transparency at the point where a parent actually needs it. C|The NYU audit of 48 platforms found that a third made no meaningful disclosure about their AI features despite those features being clearly visible in the product, and nearly three-quarters offered nothing but generic accountability and breach-response boilerplate. A company can satisfy the letter of the law and still tell a parent almost nothing useful. That gap, between legal compliance and actual transparency, is exactly where a parent's own framework has to pick up the slack.

D|What is at stake when children's data is handled carelessly

The stakes here don't end when a child closes the app. Data collected in childhood can sit in a profile for years, and hidden scoring built from that data inside advertising networks could eventually touch college admissions, job offers, loan applications, or insurance underwriting, particularly if the profile includes health or disability information pulled from an IEP or from learning-behavior records.

Schools are also, structurally, a target. Check Point data cited by DeepStrike shows the education sector logged the highest cyberattack rate of any sector worldwide in the second quarter of 2025, and Comparitech reported that ransomware groups stole millions of records from U.S. schools that same year.

A few incidents make the pattern concrete. The 2025 PowerSchool breach exposed names, addresses, contact details, and in some cases Social Security numbers and medical records. A multistate coalition of attorneys general secured a multimillion-dollar settlement in November 2025 from an education software company that failed to protect student data. Illuminate Education exposed sensitive records affecting more than 800,000 New York City students. The amended COPPA rule, which entered mandatory compliance on April 22, 2026 (per TrustArc/FTC), introduced three changes parents should know:

Voice features deserve particular attention because parents tend to underestimate them. An app that listens to a child read aloud is not logging a simple string of right and wrong answers. Under amended COPPA, the voiceprint it captures counts as biometric personal information, the same legal category as a fingerprint.

B|The counterargument deserves serious consideration, not dismissal. Limiting data collection can genuinely limit the personalization that drives the biggest learning gains, especially for disadvantaged students who benefit most from instruction that adapts to them. Privacy-conscious design has become something like a premium feature: better-funded districts can demand it and enforce it in contracts, while lower-resourced schools often end up with free-tier products that offer less protection. P|Preserving personalization does not require loosening standards. Privacy-serious design and real personalization aren't in conflict when data collection has a defined educational purpose and clear limits on retention and use.

Why privacy policies are not the right place to look

A polished privacy policy tells a parent almost nothing about what a company actually does with a child's data. F|The NYU audit found that platforms describe what they collect reasonably clearly, but say far less about what happens to that data afterward, and a family's trust in the product depends on that afterward. A policy can look thorough on page one and go silent right when it matters most.

The numbers back that up directly. A third of the platforms audited made no meaningful AI disclosure despite having visible AI features, and nearly three-quarters used only generic accountability and breach-response language, written to satisfy a legal reader rather than to inform a parent. That kind of language is written to survive a legal review, not to help a parent make a decision.

The deeper reason traces back to the same incentive problem that opened this piece. Privacy gets deferred inside edtech companies because there's little pressure from parents or schools pushing back, so a policy that keeps the company out of legal trouble gets treated as sufficient. Nobody inside the company is rewarded for writing a clearer policy, and nobody outside is demanding one loudly enough to change that calculus.

What actually predicts good behavior is not the wording of the policy but a handful of specific, checkable practices. C|Does the company say what data its AI features collect, voice, behavioral logs, error patterns, and what specific purpose that data serves, rather than hiding behind a phrase like "to improve our services"? Does it state, in the actual product agreement rather than a marketing blog post, that children's data will never be used to train AI models? Does it name its sub-processors, the SDKs, analytics tools, and A/B testing platforms that touch student data, since amended COPPA now requires every one of those third parties to be accounted for? And does it give parents a plain-language explanation of what's collected, how long it's kept, and what happens to it if the family cancels, something a non-lawyer can actually act on rather than a document written for one?

D|The questions parents should ask before trusting an AI education app with their child's data

A parent who works through a short list of concrete questions before downloading an app is in a stronger position than one who reads the privacy policy afterward. The questions below are organized by category, so they can be worked through with a specific app open on the screen.

On data collection scope: does the app record a child's voice, and if so, does the privacy policy acknowledge that voiceprints count as biometric personal information under amended COPPA and explain how long they're stored? Does it log behavioral data, reading errors, time spent on a task, wrong answers, learning pace, and does the company say what those logs are used for beyond the vague promise of "personalization"? And is data collection actually limited to what the instructional function needs, or does the app request permissions, location, contacts, the camera, that have no obvious connection to teaching a child anything?

On AI training and third-party use, the questions get sharper. Does the company commit, in its actual terms of service rather than only in marketing copy, that a child's data will never be used to train an AI model? Is there a published list of the sub-processors, the SDKs, analytics platforms, and third-party services, that touch student data, and does the company actively monitor and restrict how those sub-processors use it, as amended COPPA now expects? Does the app carry advertising or share data with ad networks at all, given that amended COPPA now requires explicit, verifiable parental consent before any such sharing for targeted advertising?

On retention and deletion: how long does the company hold onto a child's data after a subscription ends or an account is closed? Amended COPPA bars indefinite retention, so a company that can't give a specific answer to that question hasn't actually built the systems to comply with it. And can a parent request deletion of their child's data through a process that's actually described somewhere a normal person would find it?

Does the company give parents regular, plain-language progress updates that explain what the app has actually learned about their child's reading, not just a completion count? A company that can't answer that hasn't demonstrated its AI is teaching anything, which makes this as much a pedagogical question as a privacy one. Does the product avoid dark patterns, design choices built to maximize screen time regardless of whether they help a child learn, something now treated as a regulatory concern under age-appropriate design codes in California and states following its lead? H|Under the AFT/UFT/Microsoft National AI Safety & Privacy Standard, AI cannot make decisions about a student without human oversight, a principle parents can use to evaluate any app claiming to personalize instruction.

The last set of questions applies specifically when the app comes through school rather than as a parent's own choice: has the school actually reviewed the vendor's privacy policy? The Future of Privacy Forum's research found that most schools using AI edtech haven't. And does the vendor agreement explicitly rule out using student data to train AI, build marketing profiles, or share data outside the educational purpose, since the school authorization exception under COPPA never covered any of that in the first place? That last question is the natural bridge into what a genuinely strong vendor agreement looks like.

N|What Privacy-Serious Products and Agreements Look Like

Parents now have an actual public benchmark to measure a school's vendor agreement against, rather than guessing at what "good" should look like. The National AI Safety & Privacy Standard, announced September 9, 2026 by AFT President Randi Weingarten, UFT President Michael Mulgrew, and Microsoft Vice Chair and President Brad Smith, is the highest-profile example currently in place.

Its commitments are specific rather than aspirational. Student and educator data cannot be used to train AI models, sold, or used for advertising. Schools keep control over how their data is used, retained, and deleted. AI systems have to avoid manipulative or harmful design. Companies must explain, in language families can actually understand, what data gets collected and what protections apply to it. These aren't voluntary promises sitting in a press release: they're written to be incorporated directly into a district's Microsoft customer agreement, making them contractually enforceable. Starting November 1, 2026, any U.S. school district will be able to add these protections to a new or existing agreement without renegotiating the whole contract. Los Angeles Unified School District, New York State United Teachers, and New York City Public Schools, along with Mayor Zohran Mamdani, who adopted a similar framework the week before the standard was announced, have already signed on.

Legislatures are starting to build similar expectations into state law. K|A recent Idaho law directs the state education department to build a comprehensive generative AI framework, a sign that "privacy-serious" is becoming something regulators expect institutions to define and enforce.

B|At the level of an individual app, the same seriousness is visible in specific design choices rather than in policy language. Voice data captured for real-time reading feedback gets used for that feedback and nothing else, rather than quietly building a permanent voiceprint profile. K|Progress reports tell a parent what the app has actually observed about their child's specific strengths and gaps. There's no advertising, no in-app purchases, and no engagement mechanics competing with the actual goal of teaching the child something. Those are the details that separate a company that treats a child's data as a genuine responsibility from one that treats it as a line item to manage. A parent who knows what those details look like doesn't need to take a company's word for which kind it is.