Mark Cuban says Facebook lets scammers use AI-generated copies of his face and voice to sell a fake diabetes treatment, and that Meta knows it. The claim is Cuban's, made on the Pivot podcast in August and now circulating again after a Yahoo Finance article on November 5, 2026 that I could not open; the details here come from IBTimes UK and Which?. Meta has not, in the sources reviewed, answered Cuban's specific complaint. It has disputed the larger number that surrounds the story: a reported 16 billion dollar annual revenue estimate for scam and prohibited-goods ads.
This post separates what is said, by whom, and what a reader can check. It is not a ruling on Meta. For related Meta coverage, see how Meta Muse handles safety and the Facebook feed controls that launched in October.
Claimed vs verified at a glance
| Claim | Who says it | Status |
|---|---|---|
| AI copies of Cuban's face and voice sell a fake diabetes treatment | Cuban, on Pivot in August (IBTimes UK) | Cuban's account. The ads were not independently reviewed in the sources I read. |
| Buyers paid 294 dollars | Cuban, relaying messages from older Americans | Second-hand. No order data seen. |
| Some ads carry a small AI or entertainment disclaimer | Reported by IBTimes UK from Cuban's account | Reported. Cuban says it does not stop people being misled. |
| Meta projected about 10 percent of 2024 revenue, near 16 billion dollars, from scam and banned-goods ads | Internal documents reported by Reuters; repeated by Which? | Reported from leaked documents. Meta disputes it. |
| Meta "knowingly" profits | Cuban; plaintiffs in lawsuits | Allegation. No ruling cited in these sources. |
What did Cuban actually say?
According to IBTimes UK, Cuban said he regularly hears from older Americans who believe they bought a 294 dollar diabetes treatment promoted with his likeness. He says he never endorsed or sold it. The ads reportedly use AI-generated video and audio of him. Some include small disclaimers that the person shown is AI-generated or that the material is for entertainment.
Cuban argued that those disclaimers do little to stop a viewer who sees a familiar face promising a cure. He also said Facebook knows people are being targeted. IBTimes UK notes that Cuban co-founded Cost Plus Drugs, an online generic-medicine pharmacy, which makes a fake medical endorsement in his name especially damaging to his brand.
One thing worth stating plainly: the word "knowingly" is the strongest part of the claim and the hardest to verify from outside. It would require evidence that Meta had specific notice of these specific ads and chose not to act. What the public record shows is narrower, and it is about Meta's systems in general, not Cuban's ads.
Where does the 16 billion dollar number come from?
The number is not new. In November 2025, Reuters reported on internal Meta documents. According to the Which? summary and the IBTimes UK report, the documents reportedly said:
- Scam and prohibited-product ads made up about 10 percent of Meta's 2024 revenue, roughly 16 billion dollars.
- A December 2024 document estimated users were shown about 15 billion "higher risk" scam ads a day.
- Another document estimated annualized revenue from that category at about 7 billion dollars.
- A May 2025 safety presentation reportedly estimated that Meta's platforms played a role in about one-third of successful scams in the US.
- Automated systems reportedly had to be at least 95 percent sure an advertiser was committing fraud before banning the account. Below that, suspected scammers could keep advertising, reportedly at higher rates.
The 16 billion and 7 billion figures differ because they measure different things. As reported, the larger figure bundles scam ads with banned products and other violating content. The smaller one covers ads with obvious hallmarks of scams. Any headline that uses "16 billion in scam revenue" is merging the two.
What does Meta say?
Meta spokesman Andy Stone disputed the characterization. In the coverage, he called the 10 percent estimate "rough and overly inclusive" and said the documents show a selective view that distorts how the company handles fraud. Meta also said user complaints about scam ads fell by 58 percent over 18 months, and that it had removed more than 134 million scam ads in 2025 at the time of the statement. Which? reports Meta declined to give an updated revenue figure.
Meta's numbers are also self-reported. A decline in complaints measures what users report, not what runs. A count of removed ads does not say how many remain. Neither side's figure resolves the question, which is why independent audits and court discovery matter here.
Why do AI tools make this worse?
Deepfake endorsement scams are older than generative AI, but the cost of making them has collapsed. A convincing clone of a public figure's voice needs seconds of audio, and the video can be produced cheaply and swapped quickly when an ad is taken down. That lets scammers produce many variants of the same pitch, which is hard for per-ad review to catch.
The FBI's data, quoted by IBTimes UK, shows the human cost. Its Internet Crime Complaint Center recorded more than 201,000 complaints from victims over 60 in 2025, with reported losses above 7.7 billion dollars, up 59 percent from 2024. Across all ages it recorded 22,364 complaints involving AI, with adjusted losses of about 893.3 million dollars. The FBI said AI helps criminals create convincing synthetic content, including manipulated audio and video, fake profiles and personalized messages.
A fishing hook dangling a blank envelope, the lure behind a deepfake ad scam
Platform-side defenses are improving in places, such as likeness-detection tools for public figures, but how reliable they are at the scale of billions of daily ad impressions is still untested in public.
Why the 95 percent rule matters
The most important technical detail in the reporting is not the revenue. It is the enforcement threshold. A system that bans an advertiser only at 95 percent confidence will, by design, let a large number of likely scammers keep running. That is a policy choice about false positives: banning a real business by mistake is costly, so the bar is set high.
For deepfake ads the trade-off is sharp. A fake medical endorsement can look like a legitimate small advertiser for days. If the confidence score rises slowly, the campaign may finish before it is banned. Charging likely scammers more, as the documents reportedly describe, is a deterrent in theory but also raises revenue from the same ads. Regulators in several countries are asking whether that arrangement is acceptable. IBTimes UK reports the documents also indicated the US Securities and Exchange Commission was examining Meta over advertising linked to financial fraud; I could only confirm that through that single outlet.
What can you do?
- Distrust celebrity medical cures. A famous face selling a diabetes, weight-loss or cancer product is a red flag. Search the product name plus "scam", and check the person's official channels.
- Do not trust a disclaimer in the corner. A small "AI generated" or "for entertainment" label does not make the pitch honest.
- Check the checkout. Look for the seller's name, address and refund terms. Be wary of pages created recently or with only a payment form.
- Dispute fast. Contact your bank or card issuer first. In the US, also report to the FBI's IC3 portal.
- Report the ad in the app. It feeds the signals platforms use to rank risk, though results vary.
- Help older relatives set a rule. Never buy a health product from a video ad. Ask a pharmacist or doctor first.
- If you build with AI agents, treat voice and video as untrusted input. AgentBeam is the agent security platform from the explainx.ai team that stops AI agents before they take dangerous actions, such as acting on a spoofed instruction.
For the wider picture of how routine AI-assisted fraud has become, read our look at police tools that identify locations from a single photo, which shows the same technology cutting both ways, and our report on Norwegian media suing Meta.
What to watch next
Three things would move this story from allegation to fact. First, a regulator or court publishing findings on Meta's ad review, which would test the "knowingly" claim directly. Second, a Meta response to Cuban's specific ads, including whether they were removed and how quickly. Third, independent measurement: researchers sampling live ads and counting deepfake endorsements over time, which is the only way to check Meta's complaint-decline and removal figures.
Until then, the safe reading is narrow. Cuban reports deepfake ads using his likeness. Journalists report internal Meta documents with large revenue estimates. Meta says those estimates are rough. Each of those can be true at once, and none by itself proves the strongest version of the claim.
Related reading
- Is Meta's Muse safe to use? The honest verdict
- Facebook "Your Algorithm": tell Meta's AI what you want in your feed
- Meta Muse voice and live avatar video chat
- Norwegian media sue Meta in Oslo
- Meta Hatch agent changed passwords without permission
- Raven: the AI tool police use to find a house from one photo
Reports and statements are as of November 6, 2026. Allegations are not findings; check for new rulings or Meta statements.