TikTok’s Reported Safety Experiment Put Engagement Ahead of a Safeguard

Bloomberg reported that TikTok ran an experiment in which about 10% of its U.S. users—roughly 15 million people at the time—remained on an older version of the recommendation algorithm. The company was testing whether a safety update designed to reduce harmful-content clusters would make the app less engaging, or less “sticky.” 2

The report has drawn scrutiny because the control group reportedly included minors. It also included Chase Nasca, a 16-year-old whose account later received thousands of videos about suicide, sadness, hopelessness, and loneliness, according to the reporting. Nasca died by suicide weeks later. The available reporting does not establish that TikTok caused his death; it does show why the decision to withhold a safety measure has become the focus of an accountability dispute. 2

What TikTok reportedly tested

TikTok had changed its recommendation system to reduce the risk of users being overwhelmed by repeated content about high-risk subjects. The goal was to disrupt what reporting described as a harmful recommendation spiral or filter bubble, rather than allowing a user’s feed to keep reinforcing the same distressing themes. 2

Instead of deploying the change to everyone, TikTok reportedly maintained a control group of 10% of U.S. users on the previous algorithm. The stated business question was whether the safer version would reduce engagement. 2

That design matters because a conventional product test can involve a feature that is merely optional or cosmetic. Here, the disputed variable was a protection intended to limit exposure to potentially harmful content. The central ethical question is whether a platform should be able to delay such a safeguard for a randomized group when the possible downside falls on users who may already be vulnerable.

What happened to users in the control group

Users assigned to the older system could remain more exposed to repetitive recommendations involving suicide, self-harm, depression, sadness, loneliness, eating disorders, and related subjects. Reporting on Nasca’s account said it became trapped in a feed containing thousands of videos in those categories. 2

The evidence supports a careful conclusion: the experiment may have left some users without a protection TikTok had developed to reduce harmful-content clustering. It does not, by itself, prove that the algorithm caused a particular mental-health outcome or death. Distinguishing those claims is essential when assessing both the reporting and the company’s responsibility.

Why senators called the decision “depraved”

Senators Marsha Blackburn and Richard Blumenthal said TikTok knowingly withheld an algorithmic safety measure from millions of Americans—including children—to determine whether protecting users would affect engagement. They characterized the decision as “depraved” and requested a prompt explanation from the company. 1

Their concern is not limited to the outcome of one account. The senators’ questions target the decision-making process behind the test: who authorized it, which employees and executives knew about it, whether TikTok conducted a risk assessment, and why minors were allowed in the control group. 1

What TikTok is being asked to explain

The senators’ letter contains 13 questions. Among the issues it raises are:

• Why children were included in a test that withheld a safety feature.
• Which employees and executives knew about the experiment.
• Who approved the test and whether a risk assessment was completed.
• What the experiment found about safety and engagement.
• Whether TikTok ran other U.S. experiments that disabled or delayed safety features.
• Whether the company will provide the complete, unredacted internal document describing the test. 111

The requested answers are intended to establish both the scope of the experiment and the level of oversight applied to it. They could also clarify whether the reported test was an isolated product decision or part of a broader practice involving safety-related features.

What remains unproven

Several conclusions should not be drawn from the available material. The reporting does not establish that TikTok caused Nasca’s suicide, nor does it show from the provided sources how many other users experienced comparable feeds or what the experiment’s final measured effect on engagement was.

What is documented in the supplied reporting is narrower but still consequential: TikTok reportedly created a control group of roughly 15 million U.S. users, kept that group on an older recommendation system, and included minors while evaluating whether a safety change affected engagement. 2

As of TechCrunch’s report on the senators’ demand, TikTok had not immediately responded to a request for comment. 15 The requested explanations therefore focus on a basic governance question for recommendation platforms: when a company has a feature intended to reduce foreseeable harm, should engagement testing determine who receives that protection?