New York’s attorney general says TikTok built a safety button that never touched the algorithm, handed it to tens of thousands of users, and confirmed to them that it had worked. Their feeds did not change. This was not a glitch waiting on a patch. The state’s newly unsealed complaint describes it as an experiment — run on purpose, with a placebo group, and measured against how long people stayed and how much ad revenue they produced.
More than two dozen states are suing TikTok over how it designs for children. New York’s case just got sharper. The amended complaint, filed under seal in late August and ordered unsealed on October 7, adds allegations that turn a long argument about addictive design into something harder to wave away: the company’s own tool for escaping the algorithm, according to the state, was a decoy.
Letitia James — The State Says the Safety Switch Was a Prop, and It Brought the Receipts
Algo Refresh launched in March 2023. The pitch was control. A user trapped in a feed that kept serving the wrong thing could reset their recommendations and start over. In the experiments New York describes, thousands of people who asked for that reset were sorted into “ghost” or placebo groups and given a version of the tool that did nothing. Their screens said the feature was on. Their recommendations stayed put. Users outside the test received the version that actually changed things, according to Reuters and the complaint.
The count is the point, and it is not small. New York alleges TikTok ran the placebo arm repeatedly through 2023 — a group in the tens of thousands before launch, then more than 28,000 in April and more than 16,000 in May. A placebo only makes sense if you intend to compare it against a working control and study the difference. The state says the difference TikTok studied was time on the app and ad revenue. The safety feature was the variable. The users were the sample.
Scale is why this lands. The complaint notes the platform has more than nine million users in New York alone, and the setting at issue was built for people already telling the app they wanted out. This is not a fringe switch buried three menus deep. It is the emergency exit, and New York says it was painted on.
You can call a lot of things product testing. You cannot call it product testing when the product is a promise the button does not keep.
TikTok’s Well-Being Manager — One Employee Wrote Down the Problem, and the Test Ran Anyway
The complaint’s sharpest exhibit is not a metric but a person: a product manager for Digital Well Being at TikTok, who objected internally that placebo testing on Algo Refresh was wrong and that users might find it misleading. The feature, this manager wrote, was about algorithm transparency and control, and a placebo group conflicted with that purpose outright, per The Verge.
That is the whole case compressed into a few words. The employee whose job was the well-being of users flagged the contradiction, and the experiment went ahead. The state treats that as evidence of knowledge — that TikTok understood what a placebo arm meant and chose the number over the person. TikTok’s answer is that it tests features the way most companies do. A placebo arm can be argued as ordinary. A written objection from your own well-being staff is harder to file under routine.
Algo Refresh — The Reset That Only Worked for a Moment
There is a second deception layered under the first. New York alleges TikTok told users the tool would reset the algorithm behind their recommendations, when the change was only temporary. Stack that on the placebo finding and the misrepresentation escapes the test group entirely. If even the working version was oversold, every user who trusted it was told something the company, by the state’s account, did not deliver (TechCrunch).
A temporary reset is not a door out of a bad feed. It is a curtain drawn over the same room.
A 16-Year-Old in New York — The Other Experiment, and a Cost No One Recovered
The complaint reaches back to a separate set of tests, first surfaced by Bloomberg Businessweek. New York alleges TikTok withheld from roughly ten percent of U.S. users a feature meant to interrupt streams of negative content. Among those users, the state says, was a 16-year-old boy in New York who, after an onslaught of suicide and self-harm content, died by suicide in February 2022.
That is the part of the story that resists the word “experiment.” A placebo for a reset button is a lie about a feature. Switching off a guardrail for a group that includes a struggling teenager is a different category of harm, and New York stacks the two allegations on purpose: one company, one method, a range of damage that runs from a wasted tap to a funeral.
What Changes Next — A Judge Unsealed the File, and the Fraud Question Goes to Court
Three things to watch. First, whether TikTok’s internal records go public in full, because the state wants the placebo design, the revenue comparison and the staff objections in front of a jury while the company has fought to keep its files sealed or redacted across more than two dozen state cases — and this unsealing gives plaintiffs elsewhere a ruling to cite. Second, whether the line between a “ghost” test and a normal product test survives trial, since the entire defense rests on it. Third, whether regulators treat a fake control sold as safety as ordinary practice or as fraud.
Return to the person on the other end of the switch. A teenager opens an app that has worn them down, finds the one button built to give them back some say, and presses it. The screen confirms the reset. The feed returns exactly as it was: the same spiral, the same rabbit hole, the same night. The tool meant to protect them was, according to the state, doing nothing at all, and the people who built it had already been told so (The Verge).
Escape features are not decoration. They are the difference between a feed that answers to a person and a feed that owns one.
Sources: Reuters, The Verge, TechCrunch, New York Attorney General.