People visiting Substack from within the UK and Australia may find some content blurred or blocked until they complete an age verification process that uses facial age estimation and, where necessary, identity documents, the platform said in an explanatory post. According to Substack, the measures affect chat areas, direct messages, comments on posts and notes on users’ home feeds, and are being implemented in response to the respective Online Safety Acts in the two countries. [1][2]

Substack described a tiered verification flow powered by Persona’s technology: users can elect a selfie-based facial age estimation step that employs Persona’s algorithms; if that fails, they may submit a photo of a government-issued ID such as a biometric passport or driver’s licence; and a manual review route is available for those unable to complete either of the first two methods. The company said the outcome of a successful check is stored so users do not have to repeat verification. [1][2]

Persona integrates face-biometric algorithms developed by Paravision, and the partnership’s age-estimation capability scored strongly in Australia’s Age Assurance Technology Trial (AATT). The trial’s final report, published by the Australian government, assessed a range of age-assurance approaches and found Persona’s Paravision-powered solution recorded a mean absolute error (MEA) of 1.56, the best result recorded among participants, and met Technology Readiness Level 9. [1][4]

Persona highlighted those trial results in comments accompanying the announcement. “We’re proud that Persona’s solution performed at the top of its class,” Persona VP of Partnerships Sasha Dobrolioubov said, adding that the company’s platform “serves businesses across industries and powers some of the world’s most recognized platforms, including Roblox and Reddit.” Paravision’s CEO Doug Aley also emphasised investment in model performance and fairness: “These results underscore the precision and maturity of our AI models. We’ve invested deeply in accuracy and fairness because the stakes are high: age assurance must protect young users without compromising accessibility for those of proper age.” [1]

Paravision has sought additional external validation for its models. The company announced a Level 3 certification from the Age Check Certification Scheme (ACCS), the scheme’s highest designation, and separately reported perfect demographic precision in an independent ACCS bias evaluation, claims that position the vendor among a small group of providers meeting the scheme’s strictures. According to Paravision, those outcomes support the technology’s accuracy and fairness for age-restricted services. [3][5]

Government findings and independent reporting counsel caution about inevitable errors at age boundaries and the need for fallback options. The AATT final report and coverage by The Guardian noted that while effective age-assurance technologies exist, errors are more likely for users within roughly two years of critical age thresholds, and that multi-stage approaches , including ID checks and manual review , are necessary to reduce wrongful exclusion or access. Substack’s staged flow mirrors that recommendation. [4][7]

Substack has previously signalled ambivalence about the laws while committing to comply locally where required, arguing in October that the regulations “are not necessarily effective at achieving their stated aims, and they come with real costs to freedom of expression,” yet saying it will follow local legal obligations. Industry observers say platforms adopting biometric age checks must balance regulatory compliance, accuracy, user privacy and public trust as governments evaluate the technology’s role in shielding minors from harmful content. [1]

The introduction of biometric age estimation on a publisher-facing platform highlights how regulators, vendors and online services are converging on technical controls. Industry data from the AATT and ACCS certifications provide supporting evidence for vendors and platforms, but both government reports and independent journalism underscore the practical limits of AI-driven age inference and the continuing role for documentary checks and human review. [4][3][7]

📌 Reference Map:

##Reference Map:

  • [1] (Biometric Update) - Paragraph 1, Paragraph 2, Paragraph 3, Paragraph 4, Paragraph 7
  • [2] (Substack support article) - Paragraph 1, Paragraph 2
  • [4] (Australian Government AATT final report) - Paragraph 3, Paragraph 6, Paragraph 8
  • [3] (Paravision press release on ACCS Level 3) - Paragraph 5, Paragraph 8
  • [5] (Paravision ACCS bias evaluation) - Paragraph 5
  • [7] (The Guardian) - Paragraph 6, Paragraph 8

Source: Noah Wire Services