UK and EU chatbot rules for kids: three settings to copy tonight

A shield over a chat window

Lawmakers in London and Brussels have spent this year writing down, in legal language, what a chatbot should not be allowed to do to a child. Read past the politics and the list is oddly practical. Don’t remember everything the child says. Don’t pretend to be their boyfriend. Don’t be there at 2am.

None of these rules apply in Singapore. But most of the behaviour they target can already be switched off on the apps your child uses, if you know where to look. So here is what London and Brussels have decided, and how to copy the useful parts at home.

What the UK has announced

In June, the UK government announced that social media will be banned for under-16s, naming Snapchat, TikTok, YouTube, Instagram, Facebook and X. The part that got less attention was about chatbots. According to the government’s own announcement, so-called AI “romantic companion” chatbots “will have to enforce a minimum age of 18”, and “similar intimate functionalities will be restricted for under-18s on AI chatbots more widely”.

That second line matters. It means the rule follows the feature, not just the app. On that reading, a general homework chatbot that can slide into flirtation would be covered too.

The government also said it was “looking in more detail at overnight curfews and breaks in infinite scrolling for under-18-year-olds”. The plan is to bring the proposals to Parliament before Christmas, with the first protections expected in spring 2027, and Ofcom running a study on what counts as effective age checking. So: announced, not yet in force, and some pieces still under review.

What the EU has proposed

The European Commission went further on 17 September with a draft law it calls the KIDS Act. It is a proposal, not a law. It still has to get through the European Parliament and the Council, and lawyers reading it note that details may change on the way.

Even so, the chatbot section is a useful checklist. Summaries by the law firms Davis Wright Tremaine and Greenberg Traurig describe it like this:

  • AI companions must avoid designs that create emotional dependency through simulated relationships.
  • Memory of earlier chats with a child would be off by default, unless it is needed for the child’s safety.
  • Providers would have to test for risks to children before launch, and keep monitoring after.
  • Under-13s could use these services only through parental controls.
  • Fines could reach 6% of worldwide annual turnover.

Notice what is not on the list. Nobody is proposing to ban children from asking a chatbot about photosynthesis. The worry is narrower and, we think, correct: a tool that remembers a child’s secrets, plays the part of a friend, and never sleeps is a different product from one that answers a maths question and forgets it.

The three settings to copy tonight

We will use ChatGPT for the steps because OpenAI documents its settings clearly. Other assistants have their own versions, and the logic carries over.

1. Check what it remembers. On your child’s account, open Settings, then Personalization, then Memory. OpenAI’s help page says this is where you will find controls such as Saved memories and Reference chat history. Open the saved memories with your child and read them together. It is often a surprising list. You can delete single items, or choose “Delete and turn off memory” from the summary’s ••• menu. That wipes what is remembered and switches memory off, though OpenAI notes it does not delete past chats.

This is the closest thing to the EU’s “memory off by default” idea, and it is worth doing even if you decide to turn memory back on afterwards. The point is that you both know what is being kept.

2. Use Temporary Chat for anything personal. OpenAI says temporary chats “do not create or update memories, even with personalization on”. If your child wants to ask about something they would rather not have stored, like a falling-out with a friend, this is the mode for it. Teach it as a habit, not a punishment.

3. Set quiet hours. If you have linked your account to your teen’s through ChatGPT’s parental controls, go to Settings, then Parental controls, pick your teen under Family members, and turn on Quiet hours. You then choose a start and end time, and can add more than one block. During those hours your teen cannot use ChatGPT. It is off unless you turn it on.

Two honest caveats. The parental controls page lists a switch for whether ChatGPT can use saved memories, but not one that turns off every kind of memory for a teen, so step 1 still means sitting down with your child’s own settings. And linking accounts does not let you read your teen’s conversations. OpenAI is explicit about that. We covered what alerts parents do get in our piece on ChatGPT’s teen alerts.

[[photo:1]]

The companion question is the real one

The settings above are the easy part. The harder one is the thing both the UK and the EU singled out: chatbots built to feel like a relationship.

If your child uses a companion app, one where the AI has a name, a personality and a “relationship” with the user, that is the category Britain wants to put behind an age-18 gate and Europe wants to stop designing for dependency. Neither government is saying every such chat is harmful. They are saying the design deserves suspicion when the user is a child. We wrote earlier about why these apps can be hard to put down.

A Singapore parent does not need to wait for a law to ask the same question. Look at the apps on your child’s phone and sort them into two piles: tools that answer and forget, and characters that remember and befriend. The first pile mostly needs the three settings above. The second pile needs a conversation.

Disclosure: Mentus AI makes a mentoring app for children that includes AI chat, so these rules touch our own work. We have not described our product here.

Was this useful?

Sources

  1. Social media to be banned for under-16s in landmark government move to give kids their childhood back · GOV.UK, Department for Science, Innovation and Technology
  2. European Commission Proposes EU KIDS Act, a Comprehensive Child-Safety Framework · Davis Wright Tremaine
  3. EU KIDS Act: Proposed EU Rules for Online Platforms, Age Verification, and AI Chatbots · Greenberg Traurig
  4. UK announces social media ban, additional content restrictions for individuals under 16 years old · IAPP
  5. Memory in ChatGPT · OpenAI Help Center
  6. Managing parental controls in ChatGPT · OpenAI Help Center