Shufti uses on-device AI for age verification

November 15, 2026

Shufti has introduced an age-verification system that performs part of its facial analysis on the user’s device. An encrypted, one-time token sends intermediate results to a server, which completes the check. The company says the original facial image never leaves the device.

The solution is covered by US Patent No. 12,749,342 B1. The patent describes a combination of on-device facial analysis, split AI processing and One-Time Age Tokens (OTATs). A patent, however, does not confirm a product’s accuracy or its suitability for data protection requirements in every use case.

How the encrypted token works

When a user starts an age check, Shufti says the device analyses facial features and checks that a real person is in front of the camera. This is intended to make attempts to fool the system with photographs, masks or screen replays more difficult. The device then creates an encrypted token containing intermediate outputs from the AI models. The token is sent to a server, which completes the calculation and checks whether the user meets the required age threshold, such as 18 or 21.

The company says the online service receives only a pass-or-fail result, not the facial image or the user’s full identity details. Tokens are designed to be time-limited, invalidated after use and restricted to a specific merchant to make reuse more difficult. Shufti also offers other verification methods: when an estimate is uncertain or stronger evidence is required, checks against identity documents or databases can be added.

Privacy benefits do not guarantee compliance

Processing data on the device can reduce the transfer of facial images. It does not remove all data protection considerations: the local analysis still processes visual information about an individual, while the token sent to the server contains derived model outputs. Businesses need to assess what data is generated, how long it is retained, whether it can be linked to a person or service, and how it is protected against reuse.

The accuracy of age estimation also remains important. Relevant factors include error rates around the specified age threshold, potential differences between user groups and the system’s ability to detect manipulated inputs. Escalating to document checks when an estimate is uncertain may strengthen assurance, but it involves a different form of data processing that should be made clear to users.

Regulatory requirements still apply

The system is intended to give providers of age-restricted services an alternative to self-declaration and routine identity-document uploads. In the UK, Ofcom does not consider self-declaration alone a highly effective age-assurance method. Facial age estimation is among the methods that may be capable of meeting that standard. This does not mean that every product automatically meets regulatory requirements. Responsibility for effective and privacy-compliant age checks remains with the service provider.

The European Commission is also advancing privacy-preserving age-proof tools that would allow users to demonstrate eligibility without disclosing unnecessary identity information. For businesses, the assessment therefore needs to consider more than the system’s technical design: independent performance evidence, the service’s risk profile and applicable requirements all matter.

Related Articles

Operation Endgame targets malware networks

International investigators have targeted the infrastructure behind Gootloader, CastleRAT and StealC. The coordinated action focused on criminal services that provide access to compromised systems and enable follow-on attacks, including data theft and ransomware...

Digital ID does not confer signing authority

At a forum in Vienna, Austrian digital-signature provider sproof and PwC brought together banks and trust-service providers to discuss the EU Digital Identity Wallet. The consortium set up by sproof announced an open test environment where businesses can try out...

Share This