Discord replaces most face scans with an age-inference model in its global rollout

The cost line that changes here is the per-check fee. Manual verification is variable. Each face scan or ID review is a payment to a vendor and one more document sitting in someone else's store, and the store Discord used has already lost roughly 70,000 of them ^[3]. An inference model is a fixed engineering build and then close to nothing per account scored ^[4]. Discord did not say what share of accounts the model fails to classify, so neither the residual vendor bill nor the size of the remaining document pile can be worked out from the outside. That share depends on where Discord sets the model's confidence threshold.

The rest of the design rides on that default: every new and existing account starts in a teen-appropriate state that limits age-restricted servers and channels, blurs sensitive media and routes messages from unfamiliar users into a separate request inbox ^[7]. Adults who verify get age-restricted content and additional communication settings, and accounts that do not verify stay where they started ^[8]. Only age-verified adults can speak in stage channels, a carryover from the original plan ^[9].

The model still runs on personal data. It uses signals like account tenure, device information and general activity patterns across Discord communities, and Discord says private messages and message content are excluded ^[4][5]. Vishnevskiy wrote in his blog post that many people worried Discord was simply finding new ways to collect personal data, or manufacturing a problem to justify invasive solutions. He said skepticism toward the tech industry as a whole was earned. Collecting more user data was never the goal, he wrote ^[13].

From the February delay to the rollout is about seven months of rework ^[14], and the legal map filled in over that stretch. Age checks were already live in the UK and Australia under local law, and Brazil was expected to bring in similar requirements. A growing number of US states have pushed their own age verification and online safety laws ^[10][11]. Tech Times reported that this legal pressure is pushing companies like Discord toward standardising the protections globally instead of confining them to the handful of countries with existing mandates ^[16].

This is one company's policy. The case for copying it is plain enough, in that a regulator gets an age gate and the vendor-held document pile shrinks to a residue. A statute that names document or face checks as the acceptable method would push the population back through the vendor. The fixed model cost would then sit on top of the per-check fee instead of displacing it. A model that misclassifies adults into teen defaults at any volume produces a support queue and a set of users locked out of features they are entitled to. The only route out of that state is the manual check Discord was trying to avoid ^[6].