At least half of social media child safety features don’t work: study

A new study has found that the child safety features on social media platforms have a failure rate of at least 50 per cent.

Conducted by New York University and Northeastern University for the Heat Initiative and Cybersafety Research Center, the study focused on an analysis of 86 of these child-oriented features on Instagram, TikTok, Snapchat, and YouTube. After all of their tests, the researchers concluded that, on average, around 60 per cent of the features failed at their intended purpose.

The failure rates also varied greatly depending on the platform:

• Tiktok — 50 per cent
• YouTube — 55 per cent
• Instagram — 66 per cent
• Snapchat — 73 per cent

To conduct the study, researchers created dummy accounts pretending to be kids of various ages. From there, they would examine three different scenarios:

• A child who uses the app normally and comes across a safety feature
• A teen attempting to work around restrictions
• A malicious adult actor trying to bypass protections on a teen account

Throughout those scenarios, researchers evaluated child safety failure based on a variety of criteria, including whether features were “buried” (excessively hard to find and enable), “broken” (nonfunctional), “broken and buried,” or “missing” (advertised by the app maker but could not be located under the conditions it described).

Notably, the study states that some safety failures were “critical and pervasive” across all four platforms. For example, the researchers found that all of the apps promised that children would be blocked from searching for harmful content, but this wasn’t the case in practice.

On TikTok, the researchers found that searches about eating disorders and self-harm actually yielded recommendations for such related queries as “how to pretend to eat your food,” “losing yourself to mental health,” and “razor blade skin.” Instagram, meanwhile, responded to a researcher beginning to type “eating disorder” with autocomplete results featuring deliberate misspellings that pro-eating-disorder communities use to avoid the app’s blocklist.

“As researchers, we have now had the experience multiple times of reporting on failures of current tools, only to be
immediately met with promises of new tools that reportedly protect users against the same risks as those current, failing, tools,” the researchers wrote in the study’s conclusion. “What we have not seen is an honest grappling with why risks to children exist in the first place and why existing tools may be failing.”

Going forward, they hope that social media companies “can move beyond the current cycle of critical safety failures being addressed with safety theatre that quickly degrades until the next critical failure” to create “social media products that are
safe, and safety tools that are both functional and accessible.”

It’s worth noting that The New York Times reported on the study and said it was able to replicate the results found by the researchers. However, representatives from Meta, Snap and YouTube told the publication that the study included “vague claims” that “either misrepresent those features or fail to provide any examples or evidence.”

The timing of this data is particularly notable because it comes amid a global conversation about banning social media for youth. Last month, Canada officially moved forward with a bill, which followed Australia becoming the first country to implement such a ban. However, there’s been a lot of research to show that Australia’s ban hasn’t been working, with as much as 85 per cent of youth still accessing social media. There are also many privacy concerns associated with age verification features.

Experts like University of Ottawa law professor Michael Geist have also said these kinds of bans don’t hold social media companies accountable for their actual policies and content moderation. Those comments are especially relevant to this study, which shows how there are still many inherent gaps that need to be addressed with the apps themselves, ban or otherwise.

Source: Cybersafety Research Center Via: Engadget