Senators want every TikTok test that disabled a safety feature, not just the one that hit 15 million

Marsha Blackburn of Tennessee and Connecticut's Richard Blumenthal have given TikTok until 1 September to answer 13 questions about an experiment that switched off a safety feature for roughly 15 million American users ^[1][2][4]. The question that matters most is not about that experiment: the pair asked for a list of every algorithmic experiment TikTok has run in the United States where it "withheld, disabled, delayed, or reduced a safety feature", the number of users caught in each one, and how many of those were minors ^[8].

The specific test came out of a confidential 2023 TikTok document, produced in litigation against the largest social media companies and placed under seal by a court, whose contents Bloomberg Businessweek reported on 4 August ^[11]. According to the document, TikTok disabled an algorithmic safeguard designed to break up echo chambers of harmful content for 10% of its US users, making them a control group of about 15 million people at the time ^[12]. The document describes the choice as "a delicate balance across safety and the ability to measure impact on DAU (daily active users) and core metrics" ^[15]. One user in that group was Chase Nasca, a 16-year-old from Bayport, New York, selected at random on 25 January 2022, who died by suicide within a month ^[13]. Of his account, the document says: "TikTok's filter bubble prevention strategies did not take effect on this user by design" ^[14].

The rest of the letter is personnel-level: the names of every employee who was told about the experiment, an explanation of why minors were allowed into it, when executives learned of it, and a complete, unredacted copy of the internal document ^[7]. That is discovery by letter. The blanket request is something else. It treats a holdback group as a reportable event, and it asks for it retrospectively, across an entire market, broken out by age.

That is a hard thing to answer honestly. If 10% of US users came to about 15 million, TikTok's US base at the time was on the order of 150 million ^[21]. Producing the list the senators want requires that someone, at the time, recorded which feature flags were safety flags, how long each holdback ran, and which cells contained under-18 accounts. Most experimentation platforms are built to measure lift, not to preserve that kind of audit trail. Any team running production tests on minors should read the demand as a description of the logging it will be asked for later.

The legislative hook is direct. Blackburn and Blumenthal co-authored the Kids Online Safety Act, which died in the House in 2024 and returned this session; the Senate Commerce Committee voted to advance the revived version on 5 August, the day after the Businessweek story ran ^[6][18]. The Senate text keeps a duty-of-care provision requiring reasonable care to prevent mental health harms to minors, including anxiety, depression and compulsive use, and Blackburn argues it would deter future experiments by forcing safety to be designed in from the start ^[19]. The House passed its own kids online safety package in June, by a different route ^[20]. The senators also argue timing aggravates the conduct: Congress had raised concerns about the algorithm pushing young users toward harmful content since October 2021, roughly three months before Nasca was randomised into the control group ^[17][22].

TikTok did not respond to a request for comment on the letter ^[16]. For the Businessweek investigation, a spokesperson said the company is "deeply committed to the safety and well-being of users", especially teenagers, added that "our hearts break for any family that experienced a tragic loss", and pointed to continued spending on Trust and Safety detection systems and enforcement teams ^[16].

Watch whether TikTok answers the full-list question or narrows its reply to the single 2023 test, whether the sealed document is unsealed, and whether the Senate duty-of-care clause survives reconciliation with the House package.