Deepfake scam ads aren't a moderation failure at Meta — internal documents suggest they're a revenue line

Mark Cuban says he gets emails every week from people who believe they bought a diabetes cure from him for $294. "Every single week I get emails from little old men and ladies saying, 'Will my cure for diabetes work?'" the investor said on an August episode of the Pivot podcast, hosted by Kara Swisher and Scott Galloway. The ads are deepfakes built with an AI copy of his face and voice, Cuban said, and Meta runs them anyway. He called the platform "the worst of the worst." Cuban does co-found a pharmacy business, Cost Plus Drugs, which sells low-cost generics — which is precisely what makes the impersonation plausible to a victim.

His sharpest observation is about the disclaimer. Some of the ads, Cuban said, openly state that the person on screen is AI-generated and the clip is meant as entertainment. By the time anyone reads that, the pitch has landed. "And to Facebook, that's good enough," he said. A compliance artifact that arrives after persuasion is not a safeguard; it is paperwork that converts a scam into a technically-labelled scam.

The numbers behind the anecdote come from Meta's own files. On 6 November 2025, Reuters reported on internal documents showing the company expected roughly $16 billion — about 10% of its 2024 revenue — to come from ads pushing scams and prohibited products. By Meta's internal count, users across its apps were seeing some 15 billion "higher risk" ads a day: ads carrying obvious markers of fraud. One late-2024 document valued that category at about $7 billion a year. And in a May 2025 presentation, Meta's own safety staff estimated the company's platforms play a part in a third of the scams that succeed in the United States. We should be precise about what these are: internal estimates reported by Reuters, and an investor's public allegations. The material we have does not include Meta's response to Cuban's specific claims.

Taken together, though, those figures describe something other than an overwhelmed moderation team. Ten percent of revenue is not an edge case; it is a business segment. And the number reframes every enforcement delay: when removing a bad advertiser removes money, friction in the removal process is not a bug, it is an outcome the system is indifferent to at best.

What generative AI changed is the input cost. Manufacturing a convincing video of a known, trusted person used to require money, skill and time — a natural rate limiter on impersonation fraud. That limiter is gone. The cost of a persuasive fake collapsed toward zero while the cost and incentive structure of detecting and removing it stayed exactly where it was. The resulting gap is being paid for, overwhelmingly, by the people least equipped to spot a synthetic face: the elderly.

Our read: this is the short-term AI harm that actually deserves the alarm currently spent on speculative futures. Not superintelligence — industrialized fraud, running today, at measurable scale, through the largest ad system ever built. And the fix is not primarily technical. We know how to make this expensive again: financial liability for the platform on verified impersonation fraud, mandatory identity verification for advertisers using a public figure's likeness, and cryptographic provenance on advertising creative. None of that requires a research breakthrough. It requires that letting the ads run stop being the cheaper option.

The long-term picture is not hopeless — provenance standards and detection models will improve, and the same capability that generates a fake can help flag one at scale. But that only arrives if someone makes the economics point the other way. Until then, the honest summary is uncomfortable: AI did not create this fraud, it industrialized it, and the industry it is running through has, according to its own documents, priced the damage and kept going.