NYU and Northeastern Study: Half of Social Media Child Safety Features Fail Their Own Standards

What the Research Found

A study published by the Heat Initiative and the Cybersafety Research Center — authored by researchers at New York University and Northeastern University — tested 86 advertised safety features across Instagram, Snapchat, TikTok, and YouTube.

Every single platform had a failure rate of at least 50 percent.

Researchers built dummy accounts simulating children of various ages and ran three test scenarios: a child using the platform normally, a teen actively trying to circumvent a safety feature, and a malicious adult attempting to bypass protections on a separate teen account.

The bar for calling a feature a failure was concrete. A feature failed if it was too buried in privacy menus to be realistically usable, if it simply did not do what it claimed, or if it was absent from the platform entirely.

The Specific Failures

On Snapchat, adult accounts were able to "search for, find and then message the child account with zero restrictions," according to the study. That is a foundational failure. A feature designed to prevent adults from contacting minors does not prevent adults from contacting minors.

On Instagram, a teen account in the study was able to message an adult it did not follow, with no warning triggered in either direction. Meta has repeatedly promoted Instagram's "Teen Accounts" feature as a meaningful protective layer.

TikTok's algorithm suggested anorexia-related searches to accounts registered as minors. Recommending that content to a teenager is not a neutral outcome. The platform's core engagement system actively works against the safety architecture sitting on top of it.

The Platforms Pushed Back

Spokespeople for Snap, Meta, and YouTube each disputed the study's findings in statements to the New York Times. Meta's spokesperson said teens on Instagram "are seeing less sensitive content, experiencing less unwanted contact, and spending less time on Instagram at night" because of Teen Accounts.

That rebuttal deserves fair consideration. Platform-wide behavioral data, if accurate, could show genuine improvement even if specific edge-case features are inconsistently implemented. It is possible to reduce harm in aggregate while individual features remain flawed. The study tested discrete feature mechanics; Meta is pointing to population-level outcomes. Those are different measurements, and both matter.

However, the New York Times reported it was able to independently replicate the study's findings. That is a significant problem for the platforms' credibility on this specific rebuttal. The Times did not say it replicated every finding. The article does not specify which failures it confirmed. But the statement of replication is on the record.

The Broader Legal and Regulatory Context

This study lands inside a much larger set of pressures on social media companies. Platforms are currently facing lawsuits from school districts and individuals who allege their products caused direct harm to children. The litigation wave is ongoing.

Multiple countries are also moving toward outright bans. Australia went further than a ban. It has recently doubled the maximum penalty for companies found in non-compliance with its child social media restrictions, according to the report.

The U.S. has not enacted a comparable federal ban or penalty structure. Congress has debated child online safety legislation repeatedly over the past several years without passing a comprehensive law.

What Is Actually Proven, and What Is Not

The study is an academic audit, not a government enforcement action. The researchers do not allege that any platform deliberately designed its features to fail. They found the features underperform against their own stated standards.

No regulatory investigation or enforcement action tied to these specific findings has been announced as of June 29, 2026. No charges have been filed based on this study.

The platforms are corporations defending themselves against litigation and regulatory scrutiny simultaneously, which means their public statements on feature effectiveness are also interested-party statements. Neither the companies' denials nor the researchers' methodology should be accepted uncritically. But the Times replication test is the most credible independent check available in these sources.

The Open Question

The study was published by the Heat Initiative and Cybersafety Research Center — both organizations with advocacy positions on child online safety. That does not automatically invalidate the methodology, but it means independent peer review and replication by parties without a stake in the outcome would significantly strengthen or undermine the findings. Whether academic peer review or regulatory testing of these 86 features will follow is the question that determines whether this study changes anything, or just adds to a pile of reports that platforms can dispute and wait out.