Dr Deepfake: How AI Fakes Doctors to Sell Fake Medicine

Somewhere right now, someone is watching a video of a doctor they don't know personally — white coat, hospital hallway, calm reassuring voice — telling them a pill or injection changed their life. They're reaching for their wallet. The doctor never said any of it. The video is fake, built by AI, and it never happened.

That's not a hypothetical. It's what happened at UZ Gent, a major university hospital in Belgium, which put out a public warning after discovering fake videos of its own doctors — including Prof. Piet Hoebeke — circulating online, appearing to endorse products the hospital has nothing to do with, according to Belga Share. And here's the part that should actually worry you: UZ Gent isn't the first Belgian hospital to sound this alarm. UZ Leuven already warned about the same thing earlier this year, and Belgium's medical association, BVAS, flagged the trend back in April. Three separate warnings, one country, less than a year. This isn't a weird one-off. It's a pattern.

Deepfake Doctor Videos: They All Follow a Pattern

Here's where it gets interesting. Fraudsters aren't slapping fake doctor faces on ads for random junk. Research firm Check Point tracked over 200 fraudulent pharmaceutical ads since October 2025, and found that about 72% of them used fake video, cloned voices, or stolen professional identities, according to Check Point Research. The targets aren't picked at random either — diabetes, weight loss, and metabolic conditions show up again and again, which makes sense the second you think about it. Ozempic and Wegovy turned weight-loss drugs into a cultural obsession, and scammers go where the desperation and the search traffic already are. This article is part of a series — start with Voice Cloning Scams Verification Habit.

One case reported by Free Malaysia Today involved a deepfake of cardiologist Dr. Onn Akbar Ali, faked to appear endorsing herbal supplements he'd never touch, per Free Malaysia Today. This isn't confined to Belgium. NBC's TODAY separately investigated fake medical ads pushing knockoff GLP-1 drugs (that's the weight-loss/diabetes drug family Ozempic belongs to) and diabetic products across social platforms, per TODAY. Same playbook, different continent.

Why Your Brain Falls For It — Even When You're Careful

This is the uncomfortable part. Most of us were taught to trust doctors. That instinct is good — it's how medicine works. But it also means doctors are the single most effective mask a scammer could wear. Psychologists call this authority bias: when someone appears to have expertise or credentials, we skip the part of our brain that usually asks "wait, is this real?" A white coat does the persuading before a single word is spoken.

And the fakes aren't sloppy. Reporting connected to STAT News's coverage of this trend describes even trained clinicians being fooled, and notes that in one documented case, a physician's own brother-in-law believed a fake video well enough to act on it — and then recommended it to friends, some of whom became paying customers of the fake product.

Deepfakes shift the burden of trust from recognition to verification — and digital medicine's entire infrastructure assumes authenticity. — Analysis via STAT News

Translate that out of expert-speak: healthcare was never built with a "prove this video is real" step. Doctors don't watermark their appearances. Hospitals don't stamp every clip that gets posted online. There's zero friction between "this looks like my doctor" and "I trust what my doctor just told me." Scammers didn't break that system — they found the gap that was always there. Previously in this series: 3 Seconds Of Your Voice Is All A Scammer Needs To Sound Like.

Deepfake Laws Aren't Keeping Pace With the Tech

Some people will tell you not to panic because detection technology is improving fast — and it is. Intel has demonstrated deepfake detection running in real time at 96% accuracy. That sounds reassuring, right up until you look at medicine specifically. A 2026 study published in the journal Radiology found that AI-generated fake X-rays were realistic enough to fool trained radiologists — and even when those radiologists knew some of the images in front of them were fake, their average accuracy at spotting them stayed around 75%.

Sit with that number for a second. Seventy-five percent sounds like a decent grade in school. In medicine, that leaves too much room for error when your health is on the line. If radiologists — people who spend their entire careers staring at images for a living — can only catch fake medical images three out of four times, what odds does a tired parent scrolling their phone at 11pm actually have?

What You Can Actually Check Before You Trust the Video

If you've ever paused mid-scroll wondering whether the "doctor" in a video is really who the caption says — that instinct is worth trusting, not dismissing. Here's one concrete thing to do with it: before you buy anything or change any treatment based on a video, go find that doctor or hospital's real, official page directly (search for it yourself, don't click through from the ad) and check whether they've posted anything matching the claim. Real hospitals like UZ Gent don't quietly let their doctors endorse random supplements — they put out public statements when it happens without their consent, exactly like the one that sparked this story. If you can't find the same claim on the hospital's own verified channel, that silence is your answer. Up next: Your Moms Voice On The Phone Isnt Proof Anymore Heres The 10.

Ask Yourself This Before You Scroll Past

Here's the question worth sitting with: if a video appeared tomorrow showing your actual doctor — the one you've seen in person, whose handwriting you recognize on a prescription pad — recommending a treatment you'd never heard of, what would actually make you pause? Not "would you notice it's fake." Would you even think to check?

UZ Gent didn't catch this because a machine flagged it. A hospital had to publicly announce that its own doctor's face was being used without consent to sell something he never touched — and that's the system currently working as intended. The pill in that fake ad isn't the dangerous part. The fact that nobody had to fake the trust is.

People sometimes ask why nobody catches a deepfake before it spreads across social media. Part of the answer is that fake detection tools and detection models are built for one narrow job — spotting a single audio-video deepfake or a single manipulated photo — while the internet keeps producing new multimodal deepfake formats that blend voice, video, and text together in ways older detection methods were never trained on. A audio-video fake that pairs a cloned voice with a lightly altered video clip can slip past filters designed to catch only one type of manipulation at a time. That's part of why policy around this space keeps lagging behind the technology itself.

Health awareness campaigns from hospitals like UZ Gent matter because they fill a gap that detection software can't close on its own. When a hospital publicly says "this deep fake of our doctor is not real," that statement does something no algorithm currently can: it puts a verified answer directly in front of the people searching for one. Awareness, in this sense, is a form of health protection, not just a public relations move.

It also helps to understand why deep fakes of doctors work so well compared to deep fakes of, say, celebrities endorsing sunglasses. Physician identity carries a specific kind of trust that's built over years of licensing, training, and direct patient contact. When someone fakes physician identity in a video, they're not just borrowing a face — they're borrowing decades of earned credibility that took a real doctor a career to build and that a scammer can steal in an afternoon with the right software.

Researchers building fake detection systems for medical content face a harder problem than most people realize. Datasets used to train detection models for general deepfake spotting often don't include enough real-world medical footage — hospital hallways, clinical lighting, the specific cadence of a doctor speaking to a patient — so a detection method tuned for celebrity deepfakes may miss the exact style of audio-video deepfake now circulating in fake pharmaceutical ads. Building better datasets specific to health content is one of the quieter fixes experts say is still missing.

Audio is often the weakest link people forget to check. A cloned voice can now be generated from just a few seconds of a real audio clip pulled from a hospital's own promotional video or a news interview, then layered onto a completely different video of that same doctor's face. That mismatch between audio and the source clip is sometimes the only technical clue that something is wrong, even though it's nearly invisible to a casual viewer.

None of this means detection technology is useless — it means detection technology alone isn't the fix. Policy conversations happening in places like the European Union increasingly focus on requiring disclosure labels on AI-generated health content, which would give viewers a built-in signal instead of relying entirely on their own judgment or on hospitals catching every fake after the fact. Until that kind of policy becomes standard, the verification habit this article recommends — checking the hospital's own channel directly — remains the most reliable tool available to ordinary people.

The broader lesson from UZ Gent, UZ Leuven, and BVAS sounding the same alarm in one year is that health institutions are already treating this as a policy problem, not just a technology problem. They're not waiting for perfect detection models before speaking up. That's a useful signal for the rest of us: don't wait for a perfect detection method to protect you either. Build the verification habit now, before the next deep fake shows up in your own feed.