Why every app suddenly wants to check your kid’s age

If your child has been asked to scan their face to use Roblox chat, or found their Snapchat stories suddenly visible only to friends, you have met a shift that happened remarkably fast. In about a year, most of the major platforms moved from asking a child to type a birthday to actively estimating their age — and in some cases overriding what the account claims.

This is genuinely good news, and it is also widely misunderstood. It changes some things a lot, and other things not at all.

What actually changed, platform by platform

These arrived close enough together that no single announcement made the pattern obvious.

• YouTube started the sequence, not finished it: on 29 July 2025 it began rolling out machine-learning age estimation to a small set of US users, inferring age “regardless of the birthday in the account” from signals such as what a user searches for, the categories they watch and how long the account has existed.
• Instagram began testing AI in the US on 21 April 2025 to “proactively find accounts we suspect belong to teens, even if the account lists an adult birthday, and place them in Teen Account settings.” Meta extended the testing to the UK, Canada and Australia on 21 September 2025.
• Character.AI announced on 29 October 2025 that it would remove open-ended chat for under-18s “no later than November 25”, and began doing so on 24 November 2025 in the US, in two staged groups. Under-18 accounts keep the non-chat features; conversation is gone.
• Australia’s under-16 social media restrictions took effect on 10 December 2025. On 16 January 2026 eSafety reported that age-restricted platforms had removed access to 4.7 million under-16 accounts across Australia by mid-December 2025.
• Roblox began enforcing an age check before chat works at all — Australia, New Zealand and the Netherlands in the first week of December 2025, then the rest of the world wherever chat is available from early January 2026. Users are sorted into six groups: under 9, 9-12, 13-15, 16-17, 18-20 and 21+. Facial age estimation is the default route; users 13 or older can use ID verification instead.
• Snapchat began rolling out friends-only sharing for 13- to 15-year-olds on 10 June 2026: a profile visible only to mutually accepted friends, and no Spotlight content distributed to non-friends.
• Discord is the one that has not landed. Age assurance is live in the UK and Australia, but on 24 February 2026 Discord’s CTO said the global rollout was being delayed to the second half of 2026, that “over 90% of users will never need to verify their age”, and that a user who declines keeps “your account, your servers, your friends list, your DMs, and voice chat” — losing only access to age-restricted content and the ability to loosen teen-protective defaults.

None of these companies did this because they had a change of heart. They did it because regulators made the alternative more expensive. Australia is the sharpest example: eSafety is at pains to say “It’s not a ban, it’s a delay to having accounts”, with no penalties for under-16s or their parents — but platforms that fail to take reasonable steps face court-imposed fines of up to 150,000 penalty units, which eSafety currently puts at $54.6 million AUD. Add the UK Online Safety Act, the EU Digital Services Act and a queue of lawsuits, and the arithmetic changes.

Three different technologies, three different failure modes

“Age verification” is doing a lot of work as a phrase. What sits behind it varies, and so does how it goes wrong.

1. Facial age estimation — Roblox, and one of several options elsewhere. A camera estimates age from a selfie. Roblox says the images and video are processed by its vendor, Persona, and “deleted immediately after processing”. Useful; still wrong sometimes.
2. Behavioural inference — YouTube, Instagram and Discord. No camera. The platform infers age from account-level signals: what is searched for and watched, how long the account has existed, what kind of servers a user is in. Invisible to the user, which means a wrong call arrives with no obvious cause.
3. Documentary verification — the appeal route almost everywhere. Upload ID, or in some cases a credit card. Accurate, and the option most parents are least comfortable with.

Accuracy is not evenly distributed, and it is worth being careful about what the testing actually shows. The US National Institute of Standards and Technology runs the main independent benchmark, and its finding is that error “is strongly influenced by algorithm, sex, image quality, region-of-birth, age itself, and interactions between those factors”, with “no uniformly superior algorithm”. On a common visa-photo dataset the best mean absolute error was 3.1 years, down from 4.3 a decade earlier; NIST’s own results tables treat error above five years as a red flag. NIST does not measure skin tone at all — it says plainly, “We did not attempt to recover FST from our test photographs” — so claims that these systems are graded by skin tone are not coming from this benchmark. The honest summary: a system that is right on average can still be reliably wrong about particular children.

What age verification genuinely fixes

One thing, and it matters: an adult posing as a child is now much harder. The classic grooming approach — an adult creating an account claiming to be 12 and entering spaces built for 12-year-olds — runs into a real obstacle for the first time. On Roblox, an adult who declines the age check simply cannot use chat.

Second, defaults now favour safety without a parent finding a settings menu. A child who never opens Snapchat’s privacy settings still gets friends-only sharing, because Snap now defaults 13- to 15-year-olds into it. A YouTube account read as a teenager gets teen protections applied without anyone asking for them. For the large majority of families who never touch the controls, that is the single biggest practical improvement in years.

What age verification does not touch

This is the part worth being clear-eyed about, because the announcements do not dwell on it.

• Scams are unaffected. A “free Robux” link works exactly as well against an age-verified 11-year-old. Age checks say nothing about credentials, currency or fraud.
• Peer-to-peer harm is unaffected. Most cyberbullying comes from classmates, who are genuinely the age they claim. Verification confirms they belong there.
• Off-platform movement is unaffected. The common pattern — contact begins in a game and moves to a private chat elsewhere — steps outside whatever checked the age in the first place.
• Determined circumvention persists. Guides to defeating each system circulate freely, and the regulator is not treating the rules as settled. eSafety issued compulsory information-gathering notices to the ten age-restricted platforms it has publicly identified, and its March 2026 compliance update refers to “ongoing regulatory investigations”; the Commissioner’s own summary was that “although we can see some progress has been made, it’s clear there’s still a way to go.”
• Sextortion has not slowed. NCMEC reports that in 2025 it received “an average of 137 reports of financial sextortion a day – a 37% increase in reports daily compared to the year before”. Most of that year sits before the age checks above were switched on, so it is a measure of the problem they inherited, not a verdict on them — but nothing about an age check addresses it.

The privacy question, honestly

Handing a child’s face or ID to a platform to prove they are a child is an uncomfortable trade, and the discomfort is reasonable. Roblox states its vendor deletes images immediately after processing, which is the right design. But this is a large expansion in how much identity data flows to companies whose track record on data is mixed, and it is being driven by regulation rather than demand.

You have less choice than it appears — declining the check usually means losing the feature — but you can decide which platforms your child uses at all, and a facial scan requirement is a legitimate reason to decide a particular app is not worth it.

What to actually do

1. Get your child’s real age onto their accounts. This has flipped: a false adult birthday now costs protections that are switched on automatically, rather than buying freedom. Check the birthday on every account.
2. Expect wrong calls, and know the appeal route. A teenager restricted as a younger child, or an adult misjudged, is a normal outcome. The fix is usually ID-based.
3. Keep the settings. Verified age determines who can reach your child; it does not decide whether voice chat is on, whether purchases need approval, or whether the phone is in the bedroom overnight.
4. Re-check the platforms you had ruled out — and re-check the ones you were told had changed. Roblox, Snapchat and YouTube are meaningfully different from a year ago. Discord is the one to be careful about: outside the UK and Australia its age assurance had not shipped as of its own February 2026 update, so do not assume it is on.
5. Keep having the conversation. No age check tells you your child is being groomed, scammed or bullied. The thing that does is your child deciding to tell you.

For the per-platform detail, our guides to Roblox, Discord, Snapchat, Instagram and YouTube now all cover their current age checks, and the regional guide explains why your country changes what applies.