Social media ban: software age recognition apparently has major weaknesses

A few months before the social media ban for under-16s comes into force in Australia, a report commissioned by the government casts doubt on its technical feasibility. Among other things, the report examined whether providers' software solutions can correctly identify the age of users with the help of a selfie photo. While a fairly reliable recognition rate of over 95% was guaranteed for people over the age of 19, there appears to be a gray area around the age limit of 16. Young women and people with non-white skin color are particularly affected by this, according to the report.

According to the report, 59 percent of 14-year-olds were wrongly identified as being over 16 by the technology. Even among 13-year-olds, the figure was still 38.3 percent. Misidentification was even higher among women and people from indigenous groups. For the latter group, this is attributed to the fact that too little training data may have been included in the AI. For the age groups with low accuracy in recognition, it is being discussed whether an additional age check should be carried out using ID or confirmation by the parents. However, the effort involved here would be significantly higher.

A total of 60 recognition methods from 48 providers were examined more closely. The methods ranged from age verification by checking documents to face and movement-based age estimation and age inference based on existing information.

Government is satisfied

The Australian government believes that selfie-based age verification is a fast, reliable and privacy-friendly method of enforcing the ban that comes into force in December. Companies such as Meta and Alphabet will then have to prove that they are making serious efforts to verify age. Otherwise, they could face fines in the tens of millions.

The multipart government report also examined the possibility of building age checks not only into individual apps, but into the entire digital system. Instead of each website or app checking for itself how old a user is, these checks could already take effect on the smartphone, in the app store, at the internet provider or elsewhere in the system. Large companies such as Meta and Snap suggest that Apple and Google should take over age verification centrally and then pass this information on to apps securely.

Expert calls report "worrying"

However, the report shows that these ideas are not yet fully developed. Many technologies only exist as concepts and have not yet been properly tested. There are also legal problems: Who is responsible if the system fails? How do you protect children's privacy? And how do you prevent young people from simply circumventing the controls, for example by using VPN software? While the report is generally optimistic that such system-wide solutions are possible, it also warns that there is still a long way to go before they are fully operational.

Opinions differ whether the social media ban can really be implemented by the end of the year, according to the report. Australia's Communications Minister Anika Wells said, according to US news agency Reuters, that while there is no one-size-fits-all solution to age verification, there are many effective options. Justine Humphry, a media researcher at the University of Sydney who specializes in online safety, called the report's findings "worrying".

(mki)