Cuban: Facebook knowingly runs AI deepfake scam ads; docs say $16B fraud revenue

An AI deepfake scam ad is a fraudulent promotion that uses artificial intelligence to imitate a real person. It can make the person appear to speak, move, or endorse a product without permission. The goal is to borrow that person’s credibility and make the offer seem trustworthy.

The article gives Mark Cuban’s alleged diabetes cure as an example. Cuban said scammers used an AI copy of his face and voice in ads for a $294 product. He said people contacted him believing he had sold the cure, although he had never done so.

The technology matters because viewers may recognize the face and voice before questioning the message. Some ads reportedly include a disclaimer saying the person is AI-generated and the clip is entertainment. Cuban argued that the sales pitch may already have persuaded people by then. The article does not explain the specific software used to create the copies.

Meta’s internal documents estimated that roughly $16 billion flowed from ads promoting scams and prohibited products. That amount represented about 10% of the company’s 2024 revenue. The estimate shows why scam advertising is not described as a small or isolated problem in the documents.

The documents also said users across Meta’s apps saw about 15 billion “higher risk” ads each day. These ads had obvious markers of fraud. A late-2024 document valued that category at about $7 billion annually. The figures cover internal estimates and categories that may overlap in scope.

Meta spokesman Andy Stone disputed the presentation of the documents. He called the 10% estimate rough and overly inclusive, and said the documents gave a selective view of Meta’s approach. Meta also said complaints about scam ads had fallen 58% over the prior 18 months and that it removed more than 134 million scam ads in 2025.

A disclaimer does not necessarily undo the first impression created by a convincing video. Viewers may notice a familiar face, hear a familiar voice, and absorb the product claim before they carefully consider the warning. That matters because the ad’s purpose is to trigger interest or a purchase quickly.

Cuban said some fake ads included a disclaimer admitting that the person on screen was AI-generated and that the clip was meant as entertainment. Even so, he argued that the pitch had already worked by the time people reached that point. He said users later contacted him about buying his supposed diabetes cure.

The disclaimer may identify the deception, but it does not automatically remove the advertisement or refund anyone. The article does not provide evidence about how many viewers ignored the warning or how effective disclaimers generally are. It does show Cuban’s concern that platforms treated the presence of a disclaimer as sufficient.

The article gives several indicators that Meta’s platforms are connected to successful scams. In a May 2025 presentation, Meta’s own safety staff estimated that the company’s platforms played a part in one-third of scams that succeeded in the United States. That is a claim from internal company material, not an independent measurement described in the article.

The documents also counted about 15 billion “higher risk” ads each day across Meta’s apps. These ads had obvious fraud markers. Another late-2024 document estimated their value at about $7 billion a year. Together, the figures show both large exposure and substantial financial activity around risky advertising.

Meta disputed the documents’ portrayal. Andy Stone called them a selective view that distorted Meta’s approach to fraud and scams. He said user complaints had fallen 58% over 18 months and that the company removed more than 134 million scam ads in 2025. The SEC was also examining advertising linked to financial fraud.

A 95% certainty threshold is a very high internal standard for banning an advertiser. Under the documents described in the article, Meta’s automated systems must estimate that an account has at least a 95% chance of running a scam. Only then should the account be banned under that stated process.

The key mechanism is what happens below the threshold. If the system rates the suspected scammer below 95%, the advertiser is not banned, according to the documents. The article says that advertiser can simply pay more per ad and remain live. That creates a window in which questionable promotions may continue reaching users.

The documents do not explain how Meta calculates the probability or how often human reviewers intervene. Meta disputed the documents’ broader portrayal of its fraud controls. Andy Stone said they were selective and distorted Meta’s approach. Still, the threshold matters because delayed action can leave suspected scam advertising active while the system gathers more evidence.

When scam ads remain online, victims can be persuaded to buy a product or service based on a false endorsement. In the article, people contacted Cuban after ordering a supposed $294 diabetes cure he had never sold. That example shows how delayed removal can leave users believing a fraudulent offer is legitimate.

Advertisers running scams may gain more time and reach. Meta’s documents said suspected scammers below the 95% banning threshold could keep advertising by paying more per ad. If ads are removed later, the advertiser may lose access to that campaign, but the article does not describe penalties or refunds. It also does not quantify victims’ total losses.

Meta can face criticism over its enforcement choices and the revenue connected to risky ads. The company disputed the internal documents’ interpretation and reported removing more than 134 million scam ads in 2025. The Securities and Exchange Commission was investigating advertising tied to financial fraud, adding regulatory pressure.

Targeted advertising systems connect advertisers with users likely to see a promotion. The platform sells that access and earns money when advertisers pay for ads, typically based on delivery, impressions, clicks, or related results. The article does not detail Meta’s exact billing formula, but it links advertising activity to major revenue figures.

The incentive problem appears when a risky ad can still generate payment. Meta’s documents estimated that about $16 billion, or 10% of 2024 revenue, came from ads pushing scams and prohibited products. They also said suspected scammers below a 95% certainty threshold could pay more per ad and remain live.

That structure can make enforcement difficult. Removing a doubtful ad may protect users but also stops associated revenue. Keeping it online can preserve revenue until systems are sufficiently confident it is a scam. Meta said the documents were selective and overly inclusive, while reporting that complaints fell 58% and more than 134 million scam ads were removed in 2025.