The network, exposed by Spanish fact-checking outlet Maldita.es in late July and shared with Polish Press Agency (PAP), spans more than 30 languages. The Polish-language channels, several created within the last three months, use AI avatars resembling physicians and names built around words like "dr" and common surnames such as Kowalski or Nowak.
Their videos push claims with no grounding in medicine: that sunscreen causes cancer, that eggs treat cancer and diabetes, and that people with blood pressure of 180/110 mm Hg need not see a doctor.
Similar material circulates on Facebook and TikTok, where AI-generated "herbalist grandmothers" and veterinarians have promoted herbal remedies over medication - in some cases alongside paid e-books. One Facebook clip touting a leaf as an undisclosed cancer cure drew 90,000 views.
Katarzyna Lipka, of the Polish research institute NASK's disinformation-tracking unit, said the content is most dangerous for people with conditions that turn fatal without prompt treatment.
"In the case of many diseases, and cancer in particular, the effectiveness of therapy often depends on starting treatment quickly", she said, adding that videos discouraging medical care risk delaying diagnosis and worsening outcomes.
Lipka said the channels deliberately target older viewers - who are more likely to have chronic illness and, she said, may struggle more to spot AI-generated fakes - with hooks like "drink this every day after 60" or "why doctors hide these 5 tests after 65".
One account impersonating Poland's health inspectorate solicited medical histories in the comments, which Lipka said could help creators identify which health topics draw the biggest audiences.
She said 14 of the accounts reviewed overlap with a NASK database of 140 flagged channels, and that the broader pattern - beyond individual false claims - is an attempt to erode public trust in medicine and scientific institutions. TikTok removed accounts flagged by NASK, she said, but YouTube had not responded to the institute's reports.
Meta and Google, asked about the content, both pointed to policies barring deceptive health misinformation. A Meta spokesperson said accounts using deceptive practices to mislead people are not allowed and violators face action once reported. Google's Polish press office said YouTube bars content posing "a serious risk of harm" by contradicting local health authorities' guidance, applying equally to AI-generated material, with violations triggering swift enforcement.
Lipka advised viewers to watch for tell-tale signs of AI fakes: infrequent or unnatural blinking, mismatched lip movement, flat or oddly intoned voices, and stumbles over numbers or names. "If a video seems shocking, implausible or extremely controversial, that should set off a warning", she said.
She added that the EU's AI Act would help but not solve the problem, since algorithmic recommendations can repeatedly expose viewers to the same false narratives, reinforcing belief in them. Maria Dymitruk, an AI law specialist at the firm Lubasz i Wspólnicy, noted separately that the Act's deepfake-labelling duty applies mainly to businesses, not private individuals - meaning, she said, that even as companies begin over-labelling AI-generated ads, the broader flood of deepfakes online, including those built for disinformation, is likely to persist largely unchecked.
(jh)
Source: PAP