Signals · Attention
Meta charges suspected scammers more instead of banning them, and AI deepfakes are the product
Meta's documents set a 95% bar before banning a suspected scammer. Below it, the scammer pays more and stays live, and AI deepfakes fill the gap.
by Jo·3 min read·
New here? Start with the free AI Survival Kit →
Meta charges suspected scammers more instead of banning them, and AI deepfakes are the product
0:00 / 6:19
This voice is generated by AI.
Mark Cuban told the Pivot podcast in August that Facebook knowingly runs deepfake ads. The ads use an AI copy of his face and voice to sell a $294 diabetes cure he has never sold. On 6 October Yahoo Finance carried a Moneywise piece that set his claim against internal Meta documents first reported by Reuters on 6 November 2025. Those documents showed Meta expected roughly $16 billion, about 10% of its 2024 revenue, to come from ads for scams and prohibited products. Meta spokesman Andy Stone told Reuters the documents "present a selective view that distorts Meta's approach to fraud and scams."
The threshold is the business model
The most important detail in the documents is a rule. Meta's automated systems must rate an advertiser at least 95% likely to be running a scam before they ban it. Below that line, the suspected scammer pays more per ad and stays live.
That makes suspicion a pricing decision. By Meta's own internal count, users saw some 15 billion "higher risk" ads a day, and one late 2024 document valued that category at about $7 billion a year. In a May 2025 presentation, Meta's own safety staff estimated that its platforms play a part in a third of successful scams in the US.
Meta disputes the framing. Stone called the 10% estimate "rough and overly-inclusive". He said user complaints about scam ads had fallen 58% over the previous 18 months and that the company removed more than 134 million scam ads in 2025. The same documents note that the Securities and Exchange Commission is looking into Meta over advertising tied to financial fraud.
Cuban adds a detail the numbers leave out. Some of the ads carry a disclaimer saying the person on screen is AI-generated and the clip is entertainment. "And to Facebook, that's good enough," he said. By the time a viewer reaches the disclaimer, he argues, the pitch has already worked.
A familiar face is now cheap to copy
The FBI's 2025 Internet Crime Report, cited by Moneywise, records $7.7 billion in losses reported by people aged 60 and over across 201,266 complaints. That is 59% more money than the year before. AI appeared in more than 3,100 of those complaints, and those victims lost over $352 million. Those figures count only the people who filed.
This is a story about attention and who prices it. The decision to buy happens in the feed, and the owner of the feed has set a price for letting suspected fraud through. Credibility built over years can now be rendered by anyone with the tools: a founder's face, a consultant's voice, a doctor's name. AI does the routine work of copying. Telling real from fake is now the reader's job, and it falls to the people your audience trusts.
The same piece shows the delegation problem from the other side. On 26 September a Toronto man let Meta's AI agent run his Facebook Marketplace listings. It gave a buyer his street address and accepted a lowball offer. He later traced part of the problem to a permission setting he had chosen. Whoever sets an agent's permissions owns what it does.
Your Next Move
- Treat every famous face in a paid placement as unverified. That includes Cuban's. Before you act on one, search the product name with the word "scam" next to it, as Moneywise advises.
- Set a family rule this week. Work out who in your family spends the most time on Facebook and message them today. Agree that no purchase or transfer prompted by a celebrity or a familiar voice goes ahead without a call to you first. Report suspected fraud the same day at ic3.gov. The FBI says it froze $32.9 million tied to older victims' cases in 2025, and it can only chase money it knows about.
- If your face or voice carries professional weight, publish one official place to check. Use your own site or a verified profile. State what you sell and what you never sell, so clients and family have something to compare a clip against. Before you hand any task to an AI agent, read its permission settings line by line.
I build systems like the one publishing this site. → Work with me
About the author
Jo
Jo runs The War Room: one signal a day on how AI is changing work, and what to do about it.
Sources
Get the Briefing
Want more intelligence like this?
The weekly briefing, free — and the AI Survival Kit with it.
More in Signals
See allSignals · Capability
Siemens CEO Roland Busch says AI could transform society in 7 to 8 years, and AI skills will decide who gains
3 min read
Signals · Power
Windows 11 is recording which AI coding agents you run, and in one US test the upload setting was on
3 min read
Signals · Labour Market
Pay moves to the bottleneck. A careers list that is really a map of who owns the work
3 min read