Discord’s age checks will skip more than 90% of users
Discord has started sorting accounts into adult, teen and unconfirmed age groups from behavioural signals rather than ID scans, and says more than 90% of users will never be asked to prove their age.

Discord has started working out how old its users are, and it is doing it without asking most of them anything. The chat platform began sorting accounts into adult, teen and unconfirmed age groups on 23 September, part of an age assurance system its co-founder says will leave more than nine in ten people alone. Anyone placed in the adult group sees no change at all.
Discord is judging age from behaviour, not documents
Discord had promised in February to come back in the second half of the year with something less intrusive than the ID scans and facial age estimation it first proposed, after users pushed back. It has since had to launch age checks in more places, including Texas and Brazil. The version it settled on asks most accounts for nothing at all. Discord says the model reads how long an account has existed, which communities, games and servers it is attached to, how many active days it has logged and how many messages it has sent, along with system details such as the device and browser in use. It says it never reads the content of messages or calls. The result is a placement rather than a verdict, and the engineering post released alongside the launch says accounts the system cannot place confidently stay unconfirmed instead of being guessed at.
Stanislav Vishnevskiy, Discord’s co-founder and chief technology officer, argued the company is moving first to avoid something worse. “Any form of age assurance, and the possibility of sharing something biometric or an ID, is a controversial topic. Rightfully so, because of what it represents and where internet legislation around the world is heading: toward a more gated internet,” he wrote. “By launching a global approach with non-biometric options, we’re getting ahead of laws that might ask us to implement something less privacy-focused.”
What changes for teenagers
Teen accounts keep direct messages, voice calls and access to every server that is not age-restricted, and they can still talk to friends of any age. The restrictions fall on everything around that. Messages from people they are not friends with land in a message requests inbox, and switching that off requires confirming you are an adult. A warning appears before accepting a friend request from someone with no mutual friends or small shared server. Images flagged by Discord’s sensitive content filters are blurred or blocked, age-restricted servers, channels and bot commands are closed off, and full profile details and activity are visible only to friends and people in smaller servers. Accounts Discord cannot place stay unconfirmed and carry most of the same limits, with new accounts remaining unconfirmed until there is enough signal to judge them.
Five ways for an adult to confirm their age
Adults placed in the wrong group, and anyone who wants into restricted spaces, can confirm their age from user settings. The new menu includes a credit card check, where k-ID routes the details to Stripe and neither company sees the card number; Stripe keeps the last four digits, and any small temporary charge is refunded within 14 business days. The Apple App Store and Google Play can share an age range instead of a birthdate. A Google Wallet ID pass, built on a passport from Brazil, Singapore, Taiwan, the UK or the United States, will vouch for the account. AgeKey is a reusable credential that can be set up once and presented to other services. Facial age estimation is still offered, but Discord now requires it to run entirely on the device so the biometric data never leaves the phone, and every vendor must delete what it is given once the check is finished. Discord also published a directory of those vendors and first reported age assurance figures in its 2026 transparency report in May.
Our opinion
This is the rare safety rollout where the disclosure is stronger than the promise. Discord published the model’s architecture - an XGBoost classifier, the social-graph embeddings that do most of the work, the feature rankings used to audit it - on the same day as the announcement, requires facial estimation to happen on the phone, keeps card numbers away from its own servers, and has committed to deleting vendor data. It also says plainly that the model will get things wrong, and leaves accounts it cannot place in a restricted state rather than guessing. Those are the habits of a company that expects to be checked.
The part worth arguing with is what replaced the ID check. Age is now inferred from behaviour: how long you have held the account, which servers and games you frequent, how much you post, what device you use and where you sit in the social graph. Discord calls this as accurate as biometrics and hard to game, which is true in the narrow sense that nobody can fake a decade of friendships. It is still a ranking of people by inferred attributes, assembled from the same signals advertising systems use, and it hands out different rights inside one product on that basis. The engineering post is admirably technical and pointedly silent on the one number that matters, which is how often the model is wrong. Discord’s answer is that mistakes are correctable and the data will follow, and that is a fair position - but it is also the reason to keep reading the transparency reports rather than the launch post. An age estimate that is usually right is a decent way to protect teenagers. It is a poor foundation for deciding who gets to speak if the error rate never becomes public.