Cuban's deepfaked "diabetes cure" exposes the real flaw: an ad auction where fraud simply bids more

Mark Cuban says he gets emails every week from elderly people asking whether the diabetes cure they bought from him for $294 will work. He has never sold one. Speaking on an August episode of the Pivot podcast hosted by Kara Swisher and Scott Galloway, the investor said the ads are deepfakes built from an AI copy of his face and voice, that Meta runs them anyway, and that "Facebook knows the little old ladies are getting ripped off." He called the platform "the worst of the worst." Cuban does run a pharmacy business — he co-founded Cost Plus Drugs, which sells low-cost generics — which is precisely what makes the fake plausible to the people receiving it.

The accusation would be one man's complaint if the numbers did not sit underneath it. On November 6, 2025, Reuters reported on internal Meta documents indicating the company expected roughly $16 billion — about 10% of its 2024 revenue — to come from ads pushing scams and prohibited products. By Meta's own internal count, users across its apps were seeing around 15 billion "higher risk" ads per day, the ones carrying obvious fraud markers. A late-2024 document valued that category at about $7 billion a year. These are documents reported by Reuters, not findings of a court, and they describe internal estimates rather than admissions — but they are Meta's own arithmetic, not an outside critic's.

The most revealing detail is the smallest one. Cuban said some of these ads include a disclaimer stating that the person on screen is AI-generated and that the clip is meant as entertainment. By the time anyone reads it, he argued, the pitch has already landed. "And to Facebook, that's good enough." That is a compliance artifact doing exactly what compliance artifacts do when the incentive points the other way: it satisfies a policy checkbox while leaving the conversion intact. Cuban's framing — that this defines the culture of the organization — is his opinion, and we report it as such. But the mechanism he describes is not a matter of opinion; it is how disclosure rules fail everywhere, from payday lending to supplements.

Step back and the technology story is smaller than it looks. Fake celebrity endorsements are as old as patent medicine. What generative AI changed is the cost curve: producing a convincing video of a recognizable person saying something he never said now costs roughly nothing and scales to any language, any face, any product. This is the same shift we keep flagging — identity has stopped functioning as proof. A face and a voice used to be expensive to counterfeit, so we treated them as evidence. That assumption is now obsolete, and verification infrastructure has not caught up.

But the deepfake is only the weapon. The delivery system is an auction. In a performance ad market, whatever converts best can bid most, and fraud converts spectacularly precisely because it promises what honest products legally cannot — a $294 cure for a chronic disease. An auction that is neutral about truth will systematically allocate attention to the most dishonest bidder, and the enforcement cost shows up as a direct hit to the revenue line the enforcer is paid to grow. That is not a detection problem. It is an incentive problem wearing a detection problem's clothes.

Our reading: this gets fixed when fraud-linked revenue becomes a liability rather than an asset — clawbacks, victim restitution, penalties indexed to the revenue itself — and the pattern of the last year suggests it will come from ordinary consumer and competition law, not from AI-specific statutes. The optimistic case is real and we hold it: the same model class that forges the face can sign and verify provenance, and cryptographic ad identity for anyone claiming to be a real person is technically straightforward and overdue. The honest case is that we are living in the gap — generation costs collapsed years before verification arrived, and the bill is being paid by pensioners buying a cure that does not exist. That is the ugliest distribution of a transition we still believe ends in better, cheaper medicine for everyone. Closing that gap fast is not a nice-to-have; it is the price of deserving the trust the next decade of AI medicine will require.