AINEWS
Yoti handles the check by selfie, ID upload, or its own app, and sends Anthropic only a pass or fail.
Anthropic is now blocking accounts it suspects belong to people under 18 from Claude, the company's consumer chatbot. People behind those accounts are routed to a third-party age check before they can keep using the product.
Anthropic has contracted with Yoti, a UK-based identity verification vendor, to run that check. The company never sees the selfie, the ID, or any personal data from it. Yoti runs the verification, deletes the image, and returns only a pass or fail to Anthropic.
The rule itself is straightforward. Claude, the consumer product, is restricted to users 18 and over. The gating happens at account creation: new signups confirm they are 18+ to start. Anthropic is also running the check reactively on existing accounts. When the company's systems flag an account as likely belonging to a minor, the user is required to verify age to keep using Claude.
The detection is signal-based rather than identity-based. Anthropic does not look at a user's face or government ID to decide whom to flag. The system looks at activity patterns: account signals, behavior, the way the account is used. When one of those signals suggests the account holder might be under 18, the user is routed into verification. The system does not know who the user is. It only suspects.
Once flagged, the user gets three options for proving they are 18 or older. They can take a selfie for facial age estimation, which is a Yoti model reading an estimated age from the photo. They can upload a government ID document, the kind of artifact typically accepted for KYC (Know Your Customer) checks at banks and crypto exchanges. Or they can use the Yoti Digital ID app, which carries a verified "over 18" attribute derived from earlier identity checks. The support page does not say one method is required; the user picks.
According to the company's support article, Yoti is SOC2-audited. The selfie, the document image, and the personal data extracted from them are all deleted by Yoti after the check completes. Anthropic receives a binary result: pass, meaning the user is verified as 18+, or fail, meaning they are not. Anthropic does not see, store, or retain any of the underlying verification artifacts.
If the user passes, the account is reinstated. If they fail, whether because Yoti's check came back under 18 or because the user declined to complete the check, the account remains disabled. The support page does not describe an appeal path. It also does not state an effective date for the policy, how many users have been flagged so far, or what activity patterns trigger the flag. The page does not name the flagging system at all.
A Hacker News thread on the policy is largely a debate over whether signal-based detection can be evaded and whether outsourcing biometric age estimation to a single vendor is the right tradeoff. The community reaction is partial, with only the first portion of the thread visible. Anthropic has not publicly responded to specific concerns in that thread.
The policy arrives after OpenAI began estimating user age from chat behavior in 2025 and rolled out additional teen-safety rules on ChatGPT. Those are roughly a year older, and the Anthropic support page does not position this as a response.
Plenty of consumer products restrict access to people 18 and over. Few of them build the verification pipeline so the operating company never sees the underlying ID. The signal-based detection has to be precise enough to flag the right accounts, and users, once routed into the flow, have to be willing to send a selfie or an ID to a vendor they have not heard of to keep using a chatbot. Those are the two pressures this design will be tested on.