TikTok's AI Doctors Sold an FDA-Recalled Pill

Introduction

Search "health tips" on TikTok and 42 of the first 50 videos that come back show signs of AI generation. That's 84%, the highest of any term in an agency study published July 2. Some of those synthetic doctors were selling something. One of the products was Rosabella moringa capsules, sold through TikTok Shop by a New York company called Ambrosia Brands, LLC, and recalled by the FDA on February 13 for extensively drug-resistant salmonella: 10 illnesses across 8 states and 3 hospitalizations, per the agency's final outbreak update.

You didn't ask for a supplement pitch. The search returned one in a white coat. And the disclosure was there: the complaint says some of the Rosabella-linked posts carried a paid-partnership label, and at least one carried TikTok's own "Creator labeled as AI-generated" tag. The sale went through anyway.

1,198 Videos, Counted by Hand

Hallam, a UK digital marketing agency, set up a fresh TikTok account so the algorithm had no personalization history to work from, then pulled the top results for 25 search terms across health, finance, education, careers and public information. Up to 50 videos per term, 1,198 in total. Someone reviewed every one by hand for signs of AI generation or assistance: synthetic voice patterns, AI-styled captions, repetitive scripting, visual artefacts. Across all five categories, 235 videos (19.6%) showed those signs. Health was the heaviest category at 40%, and "health tips" on its own hit 84%.

Two caveats, because that figure is already getting compressed into "TikTok health tips are 84% AI." Hallam is a marketing agency, not a research institution, and this isn't peer-reviewed work. And "signs of AI generation or assistance" is a wider net than "fake doctor," since a real dietitian's clip with an AI-generated soundtrack counts the same as a fully synthetic surgeon.

The engagement split is the part I keep going back to. AI-signal videos drew fewer likes than the rest of the sample (44,500 against 69,400) and fewer comments (496 against 791), but more shares: 7,856 against 7,221, about 9% more. That gap is what keeps the content moving, and it doesn't require anyone at TikTok to have decided anything.

TikTok Shop converts the reach into revenue, through in-video purchase buttons and affiliate commissions paid on sales rather than views.

Inside the Rosabella Affiliate Program

Rosabella has been selling moringa capsules through its own site and through TikTok Shop since February 2025, per the FDA recall notice. 404 Media's Jason Koebler traced how the videos behind those sales got made.

Luca Washenko says on camera that he started Rosabella and is listed as a cofounder on trademark filings for the brand; the federal lawsuit against Ambrosia describes him as the marketing consultant the company uses to organize and coach its influencer network. In May 2025 he posted a recruitment video, "Make money promoting Rosabella products," to a group called Rosabella UGC Creators, telling prospective creators that Rosabella had paid out about $400,000 the previous month, with some individuals clearing more than $300,000. He closed with "If you like money, Rosabella is the brand for you."

Recruitment pitches inflate, so I hold that $400,000 loosely. The figure I don't is in the complaint's exhibits: a screenshot of a single transfer from Ambrosia Brands, LLC to one creator for $317,710.

Production runs in public. Harry Chang, a creator in New Zealand, titled his YouTube tutorials with the numbers, several of which came down after 404 Media asked him for comment: "This AI TikTok Shop Video Made Me $67,420 (Here's How)," "How I print $51,000/month profit with AI influencers (feels illegal)." The stack, per 404 Media: Google Veo 3, ElevenLabs voice cloning, CapCut.

The lawsuit is Human Power of N Company v. Ambrosia Brands, LLC, No. 1:26-cv-00374-ADA-ML (W.D. Tex.), filed February 17. The plaintiff sells SuperBeets, so this is one supplement company suing another under the Lanham Act, not a regulator stepping in.

It alleges Ambrosia directed the creation of hundreds of TikTok Shop videos featuring AI-generated "doctors," and alleges (on information and belief, the weakest form of federal pleading) that Ambrosia coaches creators in a private Discord. Ambrosia hasn't answered; it moved to dismiss, arguing the allegations are too thin and that "Ambrosia is fully entitled to use the features of TikTok Shop and social media, like a Discord channel, to promote its products." A footnote adds that the Shop button was already gone from the flagged posts before the case was filed, and that Ambrosia had demonetized them. The motion is undecided.

The FDA Never Found the Contamination Source

The recall went out February 13 for specific lots of Rosabella Moringa Capsules from Ambrosia Brands, LLC of New York, with 7 illnesses in 7 states counted at that point. By the outbreak investigation's final update on April 1, the total had reached 10 illnesses and 3 hospitalizations across eight states, with onsets running from September 26, 2025 to January 8, 2026. Two serotypes turned up, Newport and Kentucky, both extensively drug-resistant.

FDA ran traceback and inspections, identified that the affected moringa powder came from a single supplier, and then stated plainly that it "was not able to determine a source of contamination." The hardest physical link came out of Indiana, where the state health department collected an unopened bottle from a sick person's home; it tested positive, and whole-genome sequencing matched it to the outbreak strain.

Checking the Cures Against NCI and Moffitt

Three of the claims the Guardian documented AI doctors making are cancer claims, and the agencies that study this already answered them in writing.

Microwaving food in plastic. Moffitt Cancer Center says containers labeled microwave-safe by the FDA "are indeed safe to use in the microwave and have not been shown to cause cancer." Dioxins form when plastic burns or melts, not when it warms up, though most takeout containers aren't microwave-safe in the first place.

Deodorant and breast cancer. NCI's current fact sheet says "no scientific evidence links the use of these products to the development of breast cancer," while noting that studies have produced conflicting results. The American Cancer Society calls the evidence for a link "currently limited."

Sleeping next to your phone. NCI says cell phone frequencies "all fall in the nonionizing range" and that "the energy is too low to damage DNA." IARC's 2011 "possibly carcinogenic to humans" classification is real, and it's a hazard bucket that also holds aloe vera extract and pickled vegetables: limited evidence, not demonstrated risk, and nothing about where the phone sits at night.

Then there's Coca-Cola boiled with onion. Science Feedback and Africa Check both found no research support for it, and noted that caffeinated soda worsens dehydration during flu. Both examined the flu version; the videos pitch it for pain relief, which nobody has fact-checked separately. The Guardian also found an anti-aging product called "Hyalethinap Plus Pro Max" that it could find no evidence exists anywhere.

Who Benefits

Ambrosia Brands, LLC gets revenue. Fabricated medical authority raises what a moringa or beetroot capsule appears to be worth, and the commission structure points every incentive at conversion instead of accuracy. The product sold for a year before the recall.

The creator network gets paid off the same rail, in amounts its own members publish in their video titles.

TikTok gets money and cover. Money, because TikTok Shop takes a cut of in-app purchases, and because AI-signal content got shared about 9% more than everything else in Hallam's sample. Cover, because TikTok's answer to the Guardian was that "these findings from a marketing agency with commercial interests are not an accurate reflection of our platform," with no competing figures attached. It did remove the videos the Guardian flagged, and the accounts Science Feedback reported in February, in both cases after an outside report rather than its own detection.

Why the AI Label Didn't Help

TikTok's guidance tells creators they "must label any realistic AIGC when they upload it," and its integrity policy bars misinformation that "could cause significant harm to individuals or society," with health named alongside elections and climate.

Some of that machinery ran here. The complaint says some of Ambrosia's influencer posts carried a "paid partnership" label, and that at least one carried TikTok's own "Creator labeled as AI-generated" tag while depicting an invented person with an invented medical credential, recommending a supplement with a working Shop button underneath.

Set that against the 18 AI-doctor accounts Science Feedback turned up in a few hours in February, which impersonated three real physicians (Paul Zalzal, Garth Davis and Brad Weening) and carried no AI label at all. Those came down, but "poormaninla," the account the complaint cites for a June 14, 2025 AI-doctor post promoting Rosabella's products, was still up when 404 Media published at the end of July.

The "TikTok does nothing" version of this doesn't survive the platform's own numbers. Its self-reported Q1 2026 enforcement report, readable only through third-party summaries because TikTok's transparency page won't serve as text, puts global removals at 184 million videos, with 2.1% of them, roughly 3.86 million, taken down under the edited-media and AI-content policy. Business Insider reported this month that the system over-tags in the other direction too.

Which leaves the real hole. A label tells you how a video was made, not who is allowed to claim a medical credential inside it. And under DSHEA, the FDA can't clear a dietary supplement for safety or effectiveness before it goes on sale, which is why the only federal action in this story is a recall triggered by people who were already sick.

The Bottom Line

About one in five people use social media for health information, per a Savanta poll of more than 7,000 English adults conducted for Healthwatch England and cited in the Guardian's reporting. Healthwatch's William Pett framed the risk as people lacking "the skills to distinguish reliable information from commercial hype." The Rosabella posts are the awkward case for that: at least one arrived pre-labeled as AI-generated, others as a paid partnership, and they converted anyway.

The motion to dismiss is fully briefed and awaiting a ruling. tryrosabella.com is still taking orders. What nobody has answered is who is supposed to act on an invented doctor whose video carried the AI label the platform asks for. TikTok treats the label as the remedy. DSHEA keeps the FDA out of the marketing question until people are already in the hospital. The only party in court over any of it is a rival supplement brand suing for its share of the beet market.