A tool used to estimate age using facial analysis that is being promoted as a way to stop underage children from accessing social media or online pornography can be fooled using an aging filter on a popular photo-editing app.
Raising the minimum age for social media has public support, as governments around the world are struggling with ways to restrict children from accessing restricted content online because of the privacy concerns about uploading a government identification to prove their age.
Yoti is among the world’s best-known companies for handling online age and identity verification. Its technology is already being used by some of the world’s most popular platforms including Meta — owner of Facebook, Instagram and Threads — and has worked closely with the Australian government as it worked on age verification policy.
One of Yoti’s methods for age estimation, which Opposition Leader Peter Dutton spruiked just yesterday — is “facial analysis”, which can calculate someone’s age using a selfie. Yoti says it doesn’t retain these selfies after analysing them, promising this technology offers a way of proving someone’s age without recording their identity. Meta is already using this in Australia ahead of a global rollout.
Critics argue that age estimation through facial scans is flawed and vulnerable to being tricked. Yoti offers some measures to stop people from fooling its system, but it doesn’t change the underlying technology’s reliance on easily falsifiable information.
I know because I tricked Yoti’s age estimation into letting me “buy” a fixed-blade knife using the photograph of a 10-year-old that I had put through an aging filter on a photo-editing application.
How a Google image search and a free app fooled age verification tech
Yoti offers an online demonstration of its technology that allows anyone to test the ways people can prove their age online. The company notes its “anti-spoofing measures” aren’t enabled in the demo (more on this later).
The demo is of a fake online store that includes products including a knife, a large drone, alcohol and vapes that are restricted by age. When trying to buy them, users are prompted to confirm their age by uploading their ID, using the Yoti app or through its facial analysis technology.
I selected the selfie option. The website accessed my webcam and asked me to put my face in the camera. I found a stock image of someone who is described as a “9- to 10-year-old girl”, pulled it up on my phone and held it up to the camera. Yoti recognised the face, identified it as someone who was younger than 13, and wisely declined my knife purchase.
I then downloaded FaceApp, a free, popular photo-editing smartphone application that was launched in 2016 with millions of users, and used it to apply an “old” filter to the child’s image. Using the subsequently aged image, I went through the same process.
Except this time, Yoti said I — well, this aged-up child — was old enough to buy the knife. And in another attempt, I was able to purchase a drone.
As mentioned, Yoti has worked on how to stop people from exploiting its system. A spokesperson for Yoti, Nikki Alvey, told Crikey in an email about its various anti-spoofing measures including liveliness detection methods, which try to catch people using images or even a mask, and technologies that can detect if someone isn’t really using their selfie camera. Yoti’s customers can use these measures or choose to provide their own.
I wasn’t able to test these measures in the demo and have asked Yoti for access to a demonstration.
The company reports it is 96.99% likely to correctly identify that a child aged between six and 12 years old is under 13 years of age. Last year, Yoti reported correctly determining whether an image was of a real person 97% of the time after three attempts using its passive MyFace liveness system on a mobile phone, while incorrectly identifying a non-real image as real 1% of the time.
Even with these measures, age and identity verification may still be vulnerable to exploitation via methods like real-time deepfake face-swap technology, which is widely accessible through consumer apps like Snapchat. A 2022 report found that nine out of 10 liveliness checks were fooled by deepfakes.
Yoti’s Alvey said that bypassing its checks is extremely difficult but acknowledged it’s not impossible.
“It would require significant resources, time and expertise to beat the system. In most use cases where an age check is required, this effort would outweigh the reward,” she said.
While the industry’s attempts to stop people from circumventing their products is a constant battle, the limitations endure regarding the underlying facial analysis technology. Age estimation through facial images remains fraught. A study published this month by Australian researchers raises how these systems use indicators like wrinkles, hair and facial features, which can be altered or are not always indicative of age. This presents a risk to individuals whose age might be incorrectly estimated, restricting them from accessing things that they are legally allowed to access.
Age verification through facial analysis is sometimes presented as a way to use technology to protect kids without the privacy drawbacks, which have been the fatal flaw of previous pushes to restrict online content. In practice, the technology is not reliable enough and must still be used in tandem with old school methods. Don’t take my word for it: the other two age-verification methods from Yoti rely on providing real ID.
Any policy debate that accepts the hype and ignores the limitations of facial age estimation is doomed — unless you’re fine with 10-year-olds buying knives.
Do you have concerns about proposed age-verification technology? Let us know your thoughts by writing to [email protected]. Please include your full name to be considered for publication. We reserve the right to edit for length and clarity.
#tricked #ageverificationtoolinto #letting #child #buy #knife