New York Attorney General Letitia James has sharpened her attack on TikTok. Court records unsealed this week reveal the company ran hidden tests on thousands of young users, including children and teenagers. Those users thought they had turned on a tool called Algo Refresh to reset their video recommendations and escape harmful content. Instead, many received a non-working version. Their feeds stayed the same. TikTok watched what happened next.

The allegations come from an amended complaint in a lawsuit first filed in 2024. Reuters first reported the details on October 7. A judge ordered the updated filing released after it had sat under seal since late August. The claims paint a picture of a platform that publicly touts its commitment to youth safety while privately measuring exactly how much those safeguards might hurt engagement and advertising dollars.

Algo Refresh was supposed to give users control. Tap it, the company said, and the algorithm starts fresh. No more unwanted videos. No more loops of troubling material. But according to the lawsuit, TikTok created two groups. One received the real reset. The other got a placebo. Users in the second group saw the confirmation screen. They believed the change had taken effect. Their recommendations never budged. The difference let TikTok compare time spent in the app and ad performance between the groups.

One internal message stood out. A product manager on TikTok’s Digital Well Being team pushed back. “This feature is about algorithm transparency and control,” the manager wrote. “A placebo group completely conflicts with that.” The warning did not stop the tests. They ran throughout 2023.

The revelations build on earlier reporting. In June, Bloomberg detailed a separate experiment that withheld a safety update from roughly 15 million U.S. users starting in 2021. One teenager in the control group later died by suicide. New York now ties those tests to its broader case. “As a result, TikTok has subjected users whom TikTok included in such A/B tests to an experience that TikTok has made less safe by design,” James said in the filing.

But wait. This isn’t isolated. More than two dozen states have sued TikTok on similar grounds. They accuse the company of building an addictive product aimed at children and then misleading parents and regulators about the risks. The New York case stands out for its focus on internal decision-making and the gap between public statements and private actions.

TikTok pushed back immediately. A spokesperson told Reuters the allegations mischaracterize how Algo Refresh works. The company often tests new products and features, the statement continued, just as most technology firms do. It denied that the tests made the platform less safe for any user. In court filings, TikTok has consistently rejected the core claims of the multistate litigation.

Still, the details raise hard questions. Platforms run experiments constantly. They tweak algorithms, test new controls, measure outcomes. The difference here lies in the subject matter. When the feature promises to shield young users from harmful content, turning that promise into a control variable carries extra weight. Especially when those users include children who cannot fully grasp the stakes.

Parents have watched this story unfold for years. They hear assurances that safety teams work around the clock. They see headlines about new parental controls, screen-time limits, content filters. Then documents surface showing that some of those same tools existed only for a subset of users while others served as unwitting test subjects. The gap breeds distrust.

New York’s suit doesn’t stop at the placebo tests. It also accuses TikTok of misleading consumers about the nature of Algo Refresh itself. The company described it as a permanent reset. In reality, the effect was temporary. Users who thought they had broken free from a bad recommendation spiral soon found themselves pulled back in.

The timing of the unsealing adds pressure. Last month TikTok reached a settlement with Alabama, agreeing to pay at least $100 million and implement stricter limits on underage accounts. That deal, reported by The New York Times, could reach $300 million if the company fails to meet its obligations. Other cases continue. The pressure on ByteDance’s U.S. operations shows no sign of easing.

Industry insiders have long known that A/B testing sits at the heart of product development. Data decides which features ship and which die. Yet when the variable is youth safety, the calculus changes. Regulators now ask whether companies can ethically withhold protections from one group to prove that those protections work for another. The answer may shape rules for every major social platform.

James’s office built its case on internal documents, chat logs, and product specifications obtained through discovery. Those materials show executives weighing engagement metrics against safety outcomes. They show employees flagging ethical problems. And they show the tests continued anyway. The pattern suggests a culture that prioritized growth metrics even when those metrics conflicted with stated values.

TechCrunch noted the placebo experiment affected users who had actively sought out the safety feature. Its coverage on October 8 highlighted how the deception hit teens and children who had already signaled concern about their feeds. That detail makes the practice harder to defend as routine experimentation.

So what happens next? The New York case heads toward trial or further settlement talks. Other states watch closely. Federal lawmakers continue to debate legislation that would impose stricter design requirements on apps used by minors. And parents keep searching for tools that actually deliver the protection platforms advertise.

TikTok remains enormously popular with young audiences. Its algorithm delivers entertainment with unmatched speed and precision. That strength is also its vulnerability. The same mechanisms that keep users watching can trap them in harmful patterns. When the company experiments with the off-ramps it offers, it invites exactly the scrutiny now on display in state courthouses across the country.

The unsealed records offer a rare look behind the curtain. They show how one of the world’s most-used apps balances competing pressures: user growth, ad revenue, regulatory risk, and genuine harm reduction. In this instance, the documents suggest, the balance tipped away from the users who needed help most. Whether courts agree will determine if today’s revelations become tomorrow’s precedent.