YouTube's AI age check now runs on Singapore accounts

A shield mark standing for an automatic age check on a video account

Your child has not uploaded a selfie. They have not held an ID up to a camera, or asked you for a credit card, or typed a birthday into anything this week. And yet, if they watch YouTube on a Singapore account, Google’s software has very likely already formed a view about whether they are over or under 18.

That is not a leak or a loophole. It is the product. YouTube calls it age estimation, and Google’s own help page now lists Singapore among the countries where the model is running, alongside Australia, Brazil, Switzerland, the United Kingdom, the US and the European Economic Area.

A guess, assembled from what they watch

When YouTube announced the model, it described using AI “to interpret a variety of signals that help us to determine whether a user is over or under 18”. The signals it named were ordinary viewing behaviour: “the types of videos a user is searching for, the categories of videos they have watched, or the longevity of the account”. Mindy Brooks, Google’s vice-president for kids and families, put it the same way when the company set out its US plans, saying the model reads “signals already associated with a user’s account”.

The birthday on the account does not settle the question. The watch history does.

If the model decides a user is under 18, YouTube applies a bundle of protections without being asked. Google’s help page lists them: age-restricted videos become inaccessible, break and bedtime reminders switch on by default, recommendations stop serving repeated helpings of certain sensitive content, search results are filtered, ads go non-personalised and drop categories such as gambling and alcohol, autoplay needs a prompt, and a privacy reminder appears before commenting. Under-18 creators also get their uploads set to private by default.

If the model gets an adult wrong, the fix runs the other way. In Google’s words, “you can always verify your age by submitting your government ID, credit card or selfie through a verification process in your Google Account”.

A tablet left face up on a kitchen table
No selfie, no ID, no birthday prompt: this check happens in the background.

Why Google would rather guess than ask

There is a philosophy under this, and Google has been reasonably open about it. In late 2025, Annette Kroeber-Riel, its policy vice-president for Europe, laid out a three-step ladder: age declaration at the base, estimation in the middle, and hard verification kept for a narrow band at the top. The governing idea was that “assurance matches risk”, with lighter methods for news, education or travel and stronger checks for things like adult content or alcohol sales. At Google’s Growing Up in the Digital Age summit in March 2026, the same line came back: “The level of assurance matches the potential risk of the content or feature.”

Google’s case against checking everyone’s ID is not merely that people find it annoying. It argues a universal ID gate would “undermine privacy and exclude people without IDs”, and could create “a false sense of security, letting companies off the hook from investing in safeguards and age-appropriate experiences”.

We think that is mostly right, and it is worth saying so, because it cuts against where much of the industry is heading. Set it beside Roblox asking children for a face scan or Discord asking for an ID upload and the difference is stark: a model reading viewing categories asks nothing of the child at all. Google says the approach “does not involve collecting additional data or sharing granular user information with all apps and websites”.

The cost sits elsewhere. It sits in the fact that the check works by profiling what a person watches, and that the person being profiled never finds out unless something visibly changes. The Center for Democracy and Technology has warned that age estimation “may deprive some users from technology they should be able to access”, offering the example of an adult who watches roleplay game reviews and gets read as a child. TIME reported that a petition against the US rollout gathered more than 68,000 signatures, its authors objecting to a system that reads viewing behaviour and watch history.

What it means in a Singapore living room

Singapore did not get this by accident. Under IMDA’s Code of Practice for Online Safety for App Distribution Services, the major app stores had to start sorting under-18s, and Google said it would extend age assurance across Play, Search, YouTube, Maps and Gemini here by March 2026. The reported specifics: SafeSearch on by default for estimated minors, Maps Timeline switched off, break reminders plus limits on YouTube videos that idealise particular body weights or push social aggression, and Gemini blocked from generating images for teen users, with its answers fact-checked and carrying source citations. Ben King, Google’s managing director in Singapore, framed it as “our systems automatically providing an added layer of protection”.

Useful. Also the exact thing to watch. An automatic layer is one you cannot inspect. There is no notification to the parent, no dashboard entry reading “we think your daughter is 14”, no way to confirm the guess landed at all. And if your ten-year-old watches on your account, or on a hand-me-down phone still signed in to an adult Google account, the model has every reason to conclude it is watching an adult, because on paper it is.

A living room lit by the glow of a screen in the evening
The estimate is invisible, so parents get no notification either way.

Stop letting it guess

The ladder Google described has a step below estimation, and that step is the one parents control. A declared age beats an inferred one. If a child under 13 is on YouTube proper, Google’s own instructions are to set up a supervised kid account, where the parent picks the content setting that limits the videos and music the child can reach. Nothing then has to be worked out from their taste in cartoons.

It is a small job, and almost nobody does it while everything seems fine. MDDI surveyed 1,986 Singapore parents of children aged 2 to 17 in February 2025 and found only 37 per cent felt confident guiding their child’s digital habits, even though nearly nine in ten were already doing something about it. The gap there is not effort. It is knowing which of the many available settings is load-bearing.

This one is. Nearly everything Google now ships for under-18s hangs off the answer to a single question about your child, and in a lot of Singapore homes tonight, a model is the thing answering it.

Was this useful?

Sources

  1. Extending our built-in protections to more teens on YouTube · YouTube Official Blog
  2. Understanding age estimation on YouTube · Google (YouTube Help)
  3. Ensuring a safer online experience for U.S. kids and teens · Google
  4. A conversation on smarter and safer age assurance for Europe · Google
  5. 6 takeaways from our Growing Up in the Digital Age Summit · Google
  6. YouTube to Estimate Users' Ages Using AI · TIME
  7. New Google safeguards to bar Singapore under-18s from dating and explicit apps by 2026 · Malay Mail
  8. MDDI Study Shows Most Parents Guide Children's Digital Use But Would Like More Support · Ministry of Digital Development and Information, Singapore
  9. Set up supervised kid accounts on YouTube · Google (YouTube Help)