On September 22, 2026 (US time), Discord announced it will use an age-estimation model that infers users' ages from account activity to automatically sort users into age categories everywhere except the UK and Australia. There are three categories: a "minor" bucket for ages 13-17, an "adult" bucket for ages 18 and up, and an "undetermined" bucket for accounts lacking enough signal to classify. The rollout will happen gradually over the course of the week, and users in Japan are included. The global rollout Discord announced in February — which would have defaulted every user into minor-safe settings — was delayed after backlash. This time, the company added verification methods that don't require showing your face or an ID, and published a technical explainer of the model. That explainer describes how the system works but never states, in numbers, how often its judgments are correct. And the rollout the company touts as "getting ahead of the law" was announced only after Brazil's new law took effect and after a Texas court order — not before.
What changes if you're classified as 13-17
Nothing changes for users placed in the adult category. The changes apply to accounts in the minor category and the undetermined category. Even under the minor category, users can still message friends, use voice calls, and join servers without age restrictions as before. On top of that, protections kick in across three areas.
The first is message intake: messages from people who aren't friends get routed to message requests. If a friend request arrives from someone with no mutual friends and no shared small server, a warning is displayed. The second is content exposure: sensitive images get blurred or blocked, and users can't access age-restricted servers, channels, or bot commands. The third is visibility: profile details and activity such as the game you're currently playing are shown only to friends and members of small shared servers.
Undetermined accounts get the same protections as the minor category, except for the third one. Newly created accounts start out as undetermined until enough signal accumulates to classify them. The age category itself isn't shown on profiles and isn't visible to other members.
As for what feeds the classification, Discord says it uses signals such as how long an account has existed, the communities and servers it belongs to, the games it plays, and activity levels — and that it does not look at the content of messages or calls. The company states that "more than 90% of users won't be asked to verify their age." That figure hasn't changed since the February announcement, and no independent verification of it has been presented.
There are exceptions to the covered regions. In the UK and Australia, the age-estimation model is not currently available. Both countries already have existing age-verification systems in place since 2025, with Australia's launch dating to December 9, 2025. Japan is not among the excluded regions, so Japanese users are subject to this new automatic classification. Discord says it will roll out the minor-safety protections themselves broadly, including in regions with no legal obligation to do so.
Adults placed into the minor or undetermined category can go to "User Settings > Account Status" to verify their age. Users who have already completed age verification will not be asked again.
XGBoost and 32-dimensional embeddings disclosed — accuracy figures are not
On the same day as the announcement, Discord published a technical explainer of the age-estimation model. According to it, the model is built on XGBoost, a machine-learning method that makes predictions by stacking decision trees. The most influential feature, the company says, is a "DERE embedding" that compresses the structure of the social graph into 32 dimensions. The social graph represents the web of connections between users and communities, and the embedding converts that shape into a set of 32 numerical values.
What the model ultimately outputs is the probability that a given account belongs to an adult. Discord sets two thresholds on this probability: accounts above the upper threshold are classified as adults, accounts below the lower threshold are classified as minors, and accounts that fall in between are left unclassified. Widening the gap between the two thresholds means only accounts the model is confident about get classified, increasing the number of people left undetermined and asked to verify. Narrowing the gap reduces how many people are asked to verify, but widens the room for accounts to be misclassified. Which tradeoff is chosen depends on where the thresholds are set.
The explainer states that the ground-truth labels used for training came only from age categories confirmed via facial age estimation or ID verification across various regions between 2025 and 2026. In other words, results gathered in regions where age verification was already required became the answer key used to train this new model.
When it comes to describing performance, the technical explainer turns qualitative. It states only that, measured on a holdout evaluation set (data set aside and not used for training), the model performed at or above the effectiveness of methods widely used in regions with age-verification laws. Discord's technical explainer goes as far as explaining that the model uses XGBoost, that a 32-dimensional graph embedding is its most useful feature, and that the gap between two thresholds is left unclassified — but it does not provide numerical figures for accuracy, misclassification rate, the actual threshold values, or the share of accounts left unclassified. The size of the evaluation set is also not disclosed.
The announcement itself states that the accuracy is on par with what's achievable through biometric verification. This, too, is the company's own claim, unaccompanied by supporting figures. The one number the company has disclosed is "over 90% won't be asked to verify their age," which measures the share of people not asked to verify — a different quantity from the share of judgments that are correct. If a minor is misclassified into the adult category, that person is also counted among those not asked to verify.
The absence of disclosure does not itself mean the accuracy is low. The company says it included UK and Australia age-verification data in its May 2026 transparency report for the first time, and whether it releases comparable figures for the age-estimation model's judgments will determine whether outside parties can verify its accuracy.
Seven verification methods, and what data reaches Discord
In September 2025, a data breach hit a manual-review pathway operated by an outside customer-service vendor; according to Discord's own account, roughly 70,000 users may have had ID images exposed, while the attackers claimed a larger number. In its FAQ, the company says it has discontinued manual review through that vendor and moved manual review to k-ID.
The verification methods now offered vary by region and device, but there are seven in total. The primary contractor is k-ID, and ID scans are processed by Veratad, a k-ID partner. Based on the company's vendor disclosure page and its announcement, here's how the methods differ.
| Verification method | Processing entity | Face capture / ID submission | Age data Discord receives |
|---|---|---|---|
| Credit card | Stripe, via k-ID | Neither required | Adult, or indeterminate |
| Apple App Store | Apple (no age-verification vendor) | Neither required | Age bracket (e.g., "18+") |
| Google Play | Google (no age-verification vendor) | Neither required | Age bracket |
| Google Wallet | Google (no age-verification vendor) | Uses a passport-derived ID pass | Age bracket and ID issuing country |
| Facial age estimation | k-ID | Video selfie (processed on-device) | Estimated age |
| ID scan | Veratad (k-ID partner) | ID and selfie | Age |
| AgeKey | k-ID | Depends on the method used when it was created | Pass/fail |
Of the seven methods, five — credit card, App Store, Google Play, Google Wallet, and AgeKey — do not require capturing your face or submitting an ID on the spot. What Discord receives is a limited set of data: an age signal and the verification method used, with Google Wallet also including the ID's issuing country, according to the company. Credit card verification runs through Stripe via k-ID; App Store and Google Play verify using the age bracket already registered with each company's account; Google Wallet uses an already-configured passport-derived ID pass; and AgeKey reuses results already confirmed through another method. Per the company's own description, alongside the age data listed in the table's right-most column, the method used for verification is also passed to Discord.
Credit card verification involves a temporary charge under $0.50, refunded within 14 business days. Google Wallet uses an ID pass built from a passport, but only passports from Brazil, Singapore, Taiwan, the UK, and the US are supported. Japanese passports cannot be used for this path. AgeKey is a reusable proof of age created after completing age verification with Discord or another company, so whether a face or ID is required depends on which method was chosen for that initial verification.
Regarding facial age estimation, the company says it requires vendors to process everything on-device and requires all vendors to delete data immediately after verification. However, both the scope of the data received and the vendor-side deletion are based solely on the company's own account.
CTO and co-founder Stanislav Vishnevskiy wrote in the announcement: "Age assurance draws strong opinions and skepticism, especially when it involves sharing biometric information or a form of ID." The top three rows of the table can be read as the company's answer to that skepticism. A company that experienced an ID-image leak has now added paths that let adults prove their age using payment methods and existing app-store accounts.
From the UK's July 2025 rollout to Texas's 90-day order
This isn't the first time Discord has rolled out age verification. In the UK, the company has run age verification and minor-safe defaults since July 2025 in response to the Online Safety Act. Laying out the 14 months between that and this announcement produces the following timeline.
| Timing | Event |
|---|---|
| July 2025 | UK launches age verification and default settings in response to the Online Safety Act |
| December 9, 2025 | Australia launches minor-safe defaults and age verification |
| February 9, 2026 | Global rollout of minor-safe defaults for all users announced (planned to begin early March) |
| February 24, 2026 | Global rollout delayed to the second half of 2026 following backlash |
| March 17, 2026 | Brazil's Digital ECA (Digital ECA) takes effect |
| May 22, 2026 | Texas Attorney General sues Discord |
| July 2026 | Agreed preliminary injunction in Texas requires UK-equivalent measures within 90 days |
| September 22, 2026 | Automatic classification via the age-estimation model announced (UK and Australia excluded from the model) |
In the February 9 announcement, Discord stated it had more than 200 million monthly active users. The plan to default all of them into minor-safe settings drew backlash and was delayed 15 days later, on February 24. At the time of the delay, the company listed conditions it would meet before resuming: adding more verification methods, publishing vendor disclosures, and releasing a technical explainer, along with promises to include age-verification data in its transparency report and to introduce spoiler channels. The company says it has now met all of these conditions.
During the delay, the legal landscape moved. In Brazil, the Digital Protection Act for Children and Adolescents (Law No. 15,211/2025), known as the Digital ECA, took effect on March 17, 2026, and self-declared age verification alone is no longer permitted. In Texas, Attorney General Ken Paxton sued Discord on May 22 under, among other things, the state's Deceptive Trade Practices Act (DTPA). In July, an agreed preliminary injunction required Discord to apply UK-equivalent age verification and default settings to Texas users within 90 days and to report progress every 30 days. That order is an interim measure pending final judgment, not a ruling on the merits of illegality.
The global rollout Discord describes as getting ahead of the law was announced on September 22, 2026 — after Brazil's Digital ECA took effect (March 17, 2026) and after the Texas injunction requiring UK-equivalent measures within 90 days (July 2026). Whether the order was issued on July 1 or July 31, the number of days elapsed by September 22 is 53 to 83 — in either case, short of 90. Discord itself states in its announcement that it had no choice but to introduce age verification in Brazil and Texas.
Vishnevskiy's explanation is framed differently. The argument is that by releasing a globally consistent approach first — one that includes options not relying on biometrics — the company can get ahead of laws that might otherwise demand less privacy-conscious approaches. That argument holds for the parts of the rollout that extend minor-safety protections into regions with no legal requirement to do so.
At the same time, a timeline does not establish causation. Nothing in the record shows that the Texas order was the reason for this global rollout; what can be said is only that the announcement followed two legal demands, in that order. The European Commission also issued a recommendation on April 29, 2026, urging member states to make age-verification tools available by December 31, 2026 (not legally binding), and the number of jurisdictions issuing such demands continues to grow.
What the next transparency report will test
By Discord's own account, every condition it set at the time of the delay has now been met. But the numbers that would let users and regulators verify the company's "over 90%" claim have not yet been released.
The first thing to watch is whether the next transparency report includes figures for the age-estimation model. If any of the following appear — the share of accounts left unclassified, the number of cases where verification overturned a classification, or an estimated number of minors misclassified into the adult category — outside observers would be able to evaluate how the thresholds were set. Second is independent verification of accuracy: whether the "at or above" claim in the technical explainer can be reproduced, given the size and conditions of the evaluation set. Third is the Texas order: if the contents of the 30-day progress reports become public, it would become possible to confirm how far the UK-equivalent measures have actually been applied to Texas users.
Some practical realities for Japanese users are already clear. Discord has not disclosed the specific breakdown of verification methods shown in Japan, but Google Wallet, at minimum, cannot be used with a Japanese passport. For adults placed into the minor or undetermined category who want to avoid showing their face or an ID, the remaining options are credit card, App Store, and Google Play.
If Discord publishes its unclassified rate and misclassification rate, and outside researchers can reproduce those figures under the same conditions, users and regulators alike will finally be able to judge — using the same numbers — how far usage-based age estimation can actually substitute for showing a face or an ID.