What happened
The Australian Government's Department of Infrastructure, Transport, Regional Development, Communications, Sport and the Arts commissioned the Age Assurance Technology Trial. The Age Check Certification Scheme ran it and tested Persona's facial age estimation from 7 April to 24 June 2025. The testers sent more than 1,100 selfie portraits, plus images from schools, to Persona's system and compared its estimates with true ages. In these lab tests, the mean absolute error was less than 2 years for ages 11 to 20. The testers called this good performance for commercial systems used in retail. In school tests, the testers found acceptable accuracy, but more than 20% of students within 2 to 3 years below each gate passed it. In the lab, 25% to 36% of 18- to 20-year-olds would have needed a further verification step at an 18+ gate. The testers found slightly less consistent results for darker skin tones. They rated the product at technology readiness level 8, below Persona's own rating of 9, and wrote that further validation is needed. No mystery shopper tests were run.
Sources
-
Assessment report, by authors with no tie to the system.
Age Check Certification Scheme for the Age Assurance Technology Trial. Age Estimation Test Report: Persona. Dated 10 June 2025, version 1.2 of 21 July 2025 (archived copy). Evaluation Results, Table 2 and accompanying text.
In lab testing we observe a Mean Absolute Error of less than 2 years for ages 11 - 20, with a standard deviation of less than 1.5 years.
-
Assessment report, by authors with no tie to the system. Context or contrary evidence.
Age Check Certification Scheme. Age Assurance Technology Trial, Part D: Age Estimation. August 2025 (archived copy). Its back cover names the department that commissioned the Trial. In section D.7.3 and Figure D.4.1 the Trial team describes the trained facial analysis models that age estimation providers use.
Commissioned by the Australian Department of Infrastructure, Transport, Regional Development, Communications, Sport and the Arts, the Trial assessed 48 vendors and over 60 distinct technologies across various sectors, including social media, gaming, adult content and online retail.
Details
AI system
- Workflow
-
The testers sent face images to Persona's system and compared each estimated age with the subject's true age. They checked the results at 13+, 16+ and 18+ age gates, in the lab and in schools.
Follow-up
- Conditions for replication
-
A buffer of at least 2 to 3 years above the gate age is needed for reliable age gating. Users close to an age gate need a further age verification step.
Entry details
- Sector
-
Online safety (age assurance).
- Purpose
-
Persona built its age estimation to estimate a person's age from a single selfie for age checks.
31 more questions not yet checked
model, release, provider, role, configuration, inputs and outputs, autonomy, human oversight, AI contribution, beneficiaries, reach, distribution and access, duration, every measurement question, every money question, ownership and control, dependencies, cross-border connections, geopolitical implications, replication attempts, availability, institutional decision, grounds, action and result, success criterion, every evidence question, observed or intended, alternative assessments, related entries, other identifiers, category mappings.
Cite this record
Peter H. Vartanian. (2026). Testers find Persona age estimates within 2 years for ages 11 to 20 (AIBD-1017). Artificial Intelligence Benefits Database, edition of 1 November 2026. https://petervartanian.xyz/aibd/records/AIBD-1017.html (archived at https://doi.org/10.7910/DVN/AADSII, version 7.0)