Meta expands Instagram’s AI age checks globally without publishing error rates

By May 2026, Meta said the detection system had placed millions of accounts into protections in those four markets and was expanding on Instagram to the EU and Brazil. It also said hundreds of millions of teens had been enrolled in Teen Accounts across Instagram, Facebook and Messenger since 2024. That broader enrollment figure is not a count of teenagers discovered by AI. The rollout report dates the EU and Brazil launches to last year; Meta’s own May 2026 announcement describes their expansion at that point.

The legal stakes are substantial. A report on the announcement puts Meta’s August settlements with nearly every US state and territory at up to $18 billion, resolving lawsuits over alleged harm to teenagers. The settlements required improvements in identifying teens and children under 13; Meta links the expansion to its agreement with 52 attorneys general across states, territories and the District of Columbia.

The same report describes Meta settling last month with a teenage plaintiff in a Los Angeles trial testing allegations that Instagram and Facebook were designed to be addictive and caused mental-health problems and an eating disorder. These remain allegations, rather than findings established by this announcement.

An estimate can change settings—or remove an account

The description of the detection tools lists signals including profiles and content a user interacts with, birthday messages and visual cues in uploaded photos and videos. Meta’s May explanation describes analysis of posts, comments, biographies and captions for clues such as school grades and birthday celebrations.

Two enforcement decisions must be distinguished. Suspected users aged 13–17 can be moved into Teen Account settings. Accounts believed to belong to children under 13 can instead be deactivated. In its May explanation, Meta said those account holders would need to provide proof of age through its verification process to prevent deletion.

Meta also said users suspected of changing a birthday from under 18 to over 18 to evade protections must verify their age with an ID or facial age-estimation tools supplied by Yoti. This verification step is distinct from the proactive analysis of account activity.

For mistaken placement into Teen Account settings, Meta’s April 2025 announcement said people would have an option to change their settings. The November announcement does not explain how well that correction process works at global scale.

The visual technology also requires careful description. In a July 31 correction to its May post, Meta said its AI does not use biometric data to determine whether someone is underage, create facial fingerprints or identify the person in a photo. It said the system recognizes general visual patterns suggesting an underage person is present. Those are Meta’s descriptions, rather than independent privacy or performance findings.

In May, Meta said advanced features such as visual analysis were available only in selected countries. November’s worldwide expansion of age estimation does not establish that every detection technique is available everywhere.

Apple’s example prompt lets a user share an age range with an app without sharing an exact age or birthday. This illustrates Apple’s feature, rather than Instagram’s interface. Image: Apple. Source: Apple.

Apple supplies a different kind of age signal

Meta says it has begun using Apple’s age-range signal in Australia as an additional input. Apple operates the App Store and provides device-level family controls; its age-sharing feature builds on Child Accounts linked to a parent or guardian through Family Sharing.

In its June 2025 description, Apple said its Declared Age Range API lets developers request the age range associated with a Child Account without receiving the child’s birth date. Parents can permit sharing always, for each app request, or never, and can change that choice. Children with managed Screen Time settings cannot change the sharing choice themselves.

That is a parent-controlled account signal, not evidence in itself of an independent identity or age check. It provides an alternative to inferring age from Instagram activity, but optional sharing limits its coverage.

Australia also has a different access threshold. An Australian government announcement set December 10, 2025 as the start of a requirement for platforms including Instagram and Facebook to take reasonable steps to prevent Australians under 16 from having accounts. For that group, the regulatory objective is preventing account access, rather than simply applying teen settings.

Meta wants legislation requiring app stores to verify age and obtain parental consent, with one device signal shared consistently across apps. Its own estimation tools would serve as a backstop, feeding corrections to app stores. This would redistribute responsibility for supplying age information across the app ecosystem. Meta calls it more effective and privacy-preserving than separate platform checks, but the November announcement provides no measured comparison establishing that claim.

What would show that the checks work?

The European Commission, acting as an EU regulator under the Digital Services Act’s online-platform rules, has already challenged Meta’s underage controls. In a preliminary finding, it said Instagram and Facebook failed to diligently identify, assess and mitigate the risks of children under 13 accessing their services. False birthdays could bypass signup restrictions without effective checks, it said. That finding concerns under-13 access and is not a final ruling or an assessment of the November rollout.

Meta says the technology’s Facebook rollout is continuing, with similar technology beginning to launch globally on Threads later this year. US parents will receive information this month explaining age protections and how to help ensure their teenager’s age is recorded correctly.

The evidence needed next is more specific than a larger enrollment total: how many undisclosed teenagers the system misses, how many adults it incorrectly restricts, and how reliably users can correct errors. The November announcement supplies no numerical results for those measures. Publishing them would allow users and regulators to judge whether wider deployment also means more reliable enforcement.

Ram-Altman turns available PC memory into a parody email

Developer Iain Kay’s Linux script turns available RAM into parody emails after Sam Altman’s remarks about local AI. Its own documentation preserves Altman’s clarification—and explains why its measurements cannot test his utilization claim.