TikTok’s Safety Reset Allegedly Did Nothing for Thousands of Young Users

On October 7, 2026, Reuters reported that newly unsealed allegations in New York Attorney General Letitia James’s case accuse TikTok of repeatedly giving thousands of users, including children and teens, nonfunctional safety controls such as Algo Refresh in 2023; the company has broadly denied the state’s allegations and disputes its description of the feature. The affected users had chosen to reset their recommendations, but their For You feeds allegedly stayed unchanged while feeds in a comparison group did change.

TechCrunch’s account quotes a TikTok spokesperson saying the company “regularly tests new products and features” to understand how they work in the real world. That response describes a general testing practice; it leaves the disputed user experience at the center of the case. The tests are allegations in ongoing litigation, and the newly public material concerns past experiments rather than every reset available today.

The reset some users allegedly never received

Algo Refresh presented a deliberate choice to someone dissatisfied with the videos appearing in their For You feed. In the alleged placebo group, a person selected the control and believed the request had taken effect, yet continued receiving recommendations that had not been reset. Someone assigned the working version received the change the control was supposed to provide. The gap between those experiences is the substance of the allegation.

A placebo group can help measure whether a feature changes behavior beyond the effect of believing it has been activated. Here, that design allegedly withheld the function from people who had expressly asked for it. The tests compared groups on time spent in the app and the effect on advertising revenue. Those business measures explain why the company might study the difference; they do not establish that any particular user encountered a specific harmful video because of an Algo Refresh test.

The user’s decision matters as much as the feed’s response. People seeking a reset may be trying to escape recommendations they find repetitive, unwanted or potentially harmful. If they were assigned a version that did nothing, they could keep watching the same stream while reasonably believing they had changed its course. That is a different question from whether a functioning reset can permanently prevent familiar topics from returning.

What a working Algo Refresh is meant to do

In TikTok’s explanation of the feature, enabling a refresh initially shows For You recommendations as though the person had just joined; subsequent interactions then shape what appears, while existing settings and followed accounts remain in place. The promised fresh start is therefore an initial change in recommendations, followed by renewed personalization. It is not a promise that the feed will remain unfamiliar indefinitely.

That sequence separates the case’s two disputes about the control. One concerns whether users in the alleged placebo group received any functioning refresh when they chose it. The other concerns how lasting the change was for people who received the working version. A feed that gradually reflects new viewing habits fits the described design; a feed that never received the initial fresh start is the failure alleged in the placebo tests.

The distinction also limits what an unchanged-looking feed can establish on its own. Two videos on a familiar subject after a refresh would not reveal whether the control had failed: recommendations can change again as the person watches and interacts, and followed accounts are unaffected. Establishing that a particular account received a placebo would require evidence about its test assignment and what the feature delivered at the time.

Why the comparison matters in New York’s case

The alleged experiment differs from a routine comparison of layouts or wording because its control group was made up of people who sought a safety-related function. Both groups could believe they had taken the same action, but only one allegedly received its intended effect. That design could make a test useful for measuring engagement while leaving affected users unable to tell that the control they selected had been withheld.

The dispute is also about what users were led to expect. An internal Digital Well Being product manager reportedly objected to a placebo version because the feature concerned control over recommendations and users might find the test misleading. The warning, like the account of how the tests operated, appears in the state’s allegations; it is not a court finding about TikTok’s conduct or intent.

For families and other users, the newly public claim establishes a question about the reliability of a control during earlier tests, not a finding that today’s Algo Refresh is nonfunctional. A visible change in recommendations can suggest that a refresh acted, but a present-day feed cannot identify who was placed in a test group years earlier. Records of the assignments and what each group actually received will be central to resolving that part of the case.