Children’s data and AI: what DysAdapt does, what adults can do, and the signs to check

Artificial intelligence tools genuinely help families and teachers: reading a worksheet, reorganising it, adapting it in a few seconds. But to adapt an exercise for a child, you inevitably talk about that child. And their data deserves care.

This article explains what adapting really needs, what DysAdapt puts in place, what everyone can do on their side, and how to tell in a few minutes whether a tool looks after the data you give it.

Two kinds of data to keep to yourself

What identifies the child

Full name, date of birth, address, the name of their school and class, a photo of their face, a file number… This is identifying data. On its own or put together, it tells you which child is being talked about.

What relates to their health

"Dyslexic", "ADHD", "has an EHC plan", "sees a speech therapist": this is health data. The UK GDPR classes it as "special category" data, which needs stronger protection. And for a child, extra care matters all the more.

The good news: adapting doesn't need any of it

To adapt a worksheet, what matters is what gets in the child's way on the page: they lose their place, they tire when writing, they lose track in long instructions. These observations are enough to choose the right adaptations, and they say neither who the child is nor what their diagnosis is.

What the guidance for schools says

In England, the Department for Education has published guidance on generative AI in education. It does not ban these tools; it sets out reference points. A few useful ones:

• AI can be used in education, including to help teachers prepare and adapt resources, as long as schools stay within the law and their own policies.
• Protect personal data. Personal and sensitive information about pupils should not be entered into generative AI tools that are not covered by the school's data protection arrangements.
• AI assists, it does not replace: the adult keeps control, checks what is produced and stays responsible for it.
• Mind the age limits. Most general-purpose AI tools are not meant for young children; here it is the adults who use them.

For a teacher, this means that using AI to adapt a worksheet is perfectly possible, provided you put nothing into it that identifies the pupil or reveals their health. For a parent, this guidance does not apply at home, but it is a good benchmark.

DysAdapt was built in France, against the French Ministry of Education's equivalent framework, which says the same things: AI is allowed to adapt content to pupils' needs, with no personal data in consumer tools and the adult in charge.

What DysAdapt does

Here, concretely, is the path a worksheet takes through DysAdapt, and what is sent at each step. All of this is also detailed on our privacy page.

And the safeguards around the child:

• The child's first name is never sent to the AI. It is only there to help you find your way around your profiles. In what the model receives, the child is simply "the child".
• The profile describes needs, not a diagnosis. You are invited to tick what you observe ("loses his place when reading"…), not a medical label.
• An automatic filter removes phone numbers, email addresses, postal addresses, dates of birth and file numbers from the profile before anything is sent to the AI, in case they were typed in.
• The child has no account. The adult prepares the worksheet; the child works on the result, on paper or on screen, without ever chatting with an AI.
• No advertising, no data sold. The site's visitor statistics use a tool we host ourselves, with no cookies.
• You can delete your account and all your data at any time, from your account.

What DysAdapt cannot do for you

To be transparent: the image of the worksheet is sent as it is to read its text. If the child's name is written at the top of the sheet, it goes with the image. It is then removed from the text that was read (the Name, First name, Class fields in the header) before analysis, but the image itself has already gone. That is why the first habit below is so useful.

What adults can do

A few simple habits, with DysAdapt as with any tool:

• Hide the name before taking the photo. A sticky note, a finger, or crop the photo below the header.
• Photograph the worksheet, not the child. No faces, no exercise book open on a personal page.
• Describe what you observe, not the diagnosis. "Tires quickly when reading" is more useful for adapting, and reveals nothing medical.
• Give the profile a nickname if you prefer: "Bean" or "Year 4" works just as well.
• For teachers: if in doubt, your senior leadership team and your school's data protection officer are there to answer. Every school must have one under the UK GDPR.
• Tidy up from time to time: delete the worksheets you no longer need, and the account if you stop using the tool.

In the age of "vibe coding", a few signs to look for in a tool

Today, thanks to AI, building an app has become possible for far more people. That is a real opportunity: parents, teachers and speech therapists are building tools for needs they know from the inside. This is sometimes called "vibe coding": you describe what you want, the AI writes the code.

These tools are almost always built with the best intentions. But data protection is a profession in its own right, and it is easy to let a mistake slip through without meaning to: a database reachable without a password, an access key visible in the code, information sent to a service abroad without realising it. Nobody does it on purpose. That is exactly why it helps to know what to look at.

Here are a few simple signs that anyone can check in a few minutes before handing over information about a child: