Child safety and age assurance

1. Step 01

   Decide what each level of assurance unlocks

   Most debate about age assurance is about the method: face scans, ID documents, parental consent. Start with a different question: what should each level of confidence let a user do?

   What you know	How you know it	What it unlocks
   A typed date of birth	The user told you	Locked-down defaults: a private account, no discovery by strangers, no private messages from adults they don't know
   An age estimate	Facial or behavioral age estimation	Text and voice chat with people of about the same age
   A verified age, or a parent's sign-off	ID check, a trusted third party or verifiable parental consent	Anything that lets an adult reach a minor privately

   This puts friction where the risk is. Most users never feel it, because most features don't need more than the first row. Context and platform matter a great deal: a learning app for 8-year-olds and a game with an adult audience will draw these lines in different places.

   The lines that matter most are usually 13, 16 and 18. They're where laws change (COPPA's under-13 consent rules, minimum ages for social media, adult content) and where your product should change what's allowed. Optimize your checks for getting those boundaries right, not for precision across every age.

   Errors aren't equal. Put an adult in a teen space and they get annoyed, and some borrow a younger relative's account. Put a 15-year-old in an adult space and every protection built for them is gone. When you tune a threshold, weigh those two mistakes differently.

2. Step 02

   Set defaults that limit who can reach a child

   The protections most likely to work act before any harmful message is sent. For every account you believe belongs to a minor:

   • The account is private, and it doesn't show up in search or recommendations to adults.
   • Adults can't message, friend or voice-chat with minors they aren't already connected to.
   • Precise location is off, and photos have location data removed.
   • Gifts, currency transfers and spending are limited, especially from new accounts.
   • No advertising based on profiling. The DSA bans it when a platform is reasonably certain the user is a minor.

   The UK's Children's Code expects high-privacy defaults like these for services likely to be used by under-18s. Your own data will usually point the same way. If you look at where confirmed cases began, many start with an open message from an adult the child didn't follow. That's a default setting to change, not a moderation problem to staff up for.

3. Step 03

   Treat age as something that can change

   Most age checks happen once, at sign-up or when a feature unlocks. But accounts don't stay with one person. They get shared, sold and handed down to a younger sibling, and months later the person using the account no longer matches the age on file.

   Decide what triggers a fresh look: a report that the user is under age, behavior that doesn't fit the age on file, a long-dormant account coming back, or signs that an account has been sold. When something looks off, you have three choices:

   • Ban it. This hits many legitimate users and teaches people to hide their age better.
   • Do nothing. This leaves the gap open.
   • Put the account back on safer defaults right away and ask for a stronger check. The child stays protected while it gets sorted out, and an adult can clear it in minutes.

   The third is almost always right.

4. Step 04

   Detect grooming as a pattern

   Grooming rarely shows up in a single message. It builds over weeks. An adult account friends many younger players it has no connection to. It gifts currency early, asks about parents and whether the child is alone, moves the conversation to private voice, and eventually pushes for another app. Each step can be innocent on its own. Together they're a pattern.

   So the response should build as signals stack:

   Signals on an adult-to-minor contact	Response
   One signal	Log it. Nothing visible happens.
   Two or more within a short window	Add friction: gifting limits, a safety prompt to the child, no new private channels between the two accounts.
   A request to move to another app, or for images, after that	Restrict contact, and send the full history to a trained reviewer.

   Thresholds should come from data, not instinct. Test them against past confirmed cases and past false alarms. Set them where precision holds and the review team can keep pace. Revisit them every quarter, because offenders learn what triggers friction and route around it. Detection tools now exist to help, such as Thorn's conversation classifiers, Roblox's open-sourced Sentinel and specialist vendors, and which fits depends on your volume and how much engineering you can put behind it. A tool still needs your thresholds and your reviewers behind it.

   Privacy sets the other boundary. Limit pattern detection to adult-to-minor pairs. Start with metadata (who contacts whom, how often, from how new an account) before reading message content. Agree retention limits with Legal before launch. Check each market: rules on scanning private messages vary, especially in the EU.

5. Step 05

   Work the queue by risk

   There will be weeks when flagged relationships outnumber reviewers. When that happens, protective actions stay automatic, and the queue is worked by risk: requests to move off-platform first, then cases involving children under 13. A restricted account waiting a day for review is a cost worth paying.

   The reviewers who handle these cases need specific training, a restricted escalation path, and limits on how much of this material they see in a day. See chapter 14.

6. Step 06

   Handle the account, not just the content

   When you find child sexual abuse material or an adult enticing a child, some steps aren't optional. The content comes down, gets preserved and gets reported. In the US that means the NCMEC CyberTipline, under 18 U.S.C. § 2258A. Since the REPORT Act, that includes enticement and child sex trafficking, and you must preserve what you reported for a year. A child in immediate danger comes before everything else: escalate to law enforcement straight away. Chapter 12 covers the first hour.

   The judgment call is the account. You can act immediately and visibly, or restrict the account quietly while investigators map linked accounts and law enforcement decides what it needs. Immature programs default to the ban because it's the only lever their tooling gives them, and it looks like success on a dashboard. In practice, the offender comes back on a fresh account with none of the history that flagged them. Their other accounts go unexamined. And the child is still reachable wherever the conversation already moved.

   Offenders work across services, so one platform's ban is one move in a longer sequence. The Tech Coalition's Lantern program lets participating companies share signals about online child sexual exploitation and abuse across platforms. If you're eligible, join. If you aren't yet, design your investigation records so you could share well-structured signals when you are.

   The standard isn't speed. It's a safe child and a fully mapped network.

7. Step 07

   Give the thresholds one owner, and write every change down

   Where the age lines and detection thresholds sit affects product, safety, legal and privacy at once. If nobody owns them, each team nudges them toward its own goals and nobody notices the drift. Give them one owner and the same sign-off as a policy change.

   Write every change down, with the numbers behind it. Every threshold gets tested the day a regulator, a parent or a court asks why a specific account was or wasn't restricted. That question is only answerable if one person owned the thresholds, each change was recorded, and the decision to report to law enforcement was made by people trained for it.

8. Step 08

   Measure prevention, not just removal

   Completion rate (how many users finished an age check) is a weak metric on its own. These tell you whether the system works:

   • Adults in teen spaces: how many accounts in teen groups turn out to be adults.
   • Children in adult spaces: how many children end up in adult groups or adult-only products.
   • Time to catch an account that changed hands.
   • Time from first signal to protective action on adult-to-minor contact.
   • Share of confirmed cases caught before the move to another app.
   • Precision at each step of the grooming response, and how long high-risk cases wait for review.
   • Confirmed unsafe-contact cases per 10,000 young users, split by how the contact started, so you can close the path, not just ban the account.

1. 01

   Checking age once, at sign-up

   Re-check when an account seems to have changed hands, and fall back to safer defaults while you do.

2. 02

   Choosing a method before deciding what it unlocks

   Decide what's at stake for each feature first, then pick the lightest check that's strong enough.

3. 03

   Banning every account that looks young

   You'll hit legitimate users and teach children to lie better. Move the account to safer defaults and ask for a stronger check.

4. 04

   Reviewing messages one at a time

   Grooming is spread across weeks of contact. Look at the relationship.

5. 05

   Treating the ban as the end of the case

   Preserve, report, and map linked accounts before you tip off the offender.

6. 06

   Reporting completion rates as success

   Report who ended up where, and how much contact you prevented.

7. 07

   Letting thresholds drift

   One owner, a written change log, and the same sign-off as policy.

From Steven's writing · 7 posts

1. Oct 3, 2026Reading roundup

   Saturday reading: child safety gaps in games, Ofcom in court and California's new laws

   Australia's eSafety found Fortnite and Minecraft still mostly rely on self-declared age. Age assurance is the foundation, because every other safeguard assumes you know who's a kid.

   Also in: 16. Regulation and compliance

2. Sep 30, 2026Share

   Why early grooming detection matters

   Tools that help platforms spot grooming early protect kids and the communities around them.

3. Sep 30, 2026Essay

   Grooming is a pattern, not a message

   Responses build as signals stack on adult-to-minor contact, with thresholds tested on past cases, a queue worked by risk, privacy limits agreed up front, and four numbers that show it works.

   Also in: 5. Detection and prevention, 7. Standing up review operations

4. Sep 29, 2026Share

   Nudify apps: harm no single platform sees in full

   The image is made on one service, the tool promoted on another and the harm lands on a third. That's why cross-platform signal sharing through Lantern matters.

   Also in: 5. Detection and prevention

5. Sep 29, 2026Essay

   Age assurance: what should each check unlock?

   The method matters less than what each level of assurance lets a user do. Errors aren't equal, accounts change hands, and someone has to own the thresholds.

   Also in: 4. Writing policy and an enforcement ladder

Outside sources

• NCMEC CyberTiplinewhere US platforms report child sexual exploitation, including enticement. Its annual data shows what gets reported.
• Tech Coalition: Lanterncross-platform signal sharing about online child sexual exploitation and abuse.
• European Commission: guidelines on the protection of minorswhat DSA Article 28 expects on age assurance and defaults.
• ICO: the Children's Codethe UK's standards for services likely to be used by children.
• FTC: the COPPA RuleUS duties for services that collect data from children under 13.
• Thorn: Youth Perspectives on Online Safetywhat 9-to-17-year-olds say about their lives online.

1. Oct 4, 2026
   RevisedWith 18 other chapters

   Practical advice that varies by platform, such as cadences, sample sizes, targets and who owns what, is now set out as options with examples, so each team can choose what fits.

2. Oct 3, 2026
   Drafted

   First full draft: what each level of age assurance should unlock, safe defaults, grooming responses that build as signals stack, the account decision, and how to measure prevention.

3. Oct 3, 2026
   AddedWith 14 other chapters

   Linked the first 16 posts to the chapters they inform, and set out the ten principles behind the handbook.