AI Deepfakes: The Scam, Propaganda and Fraud Playbook

The Threat Isn't the Hollywood Deepfake. It's the Cheap One.

Forget the hyper-realistic Tom Cruise deepfake that went viral a few years ago. That required broadcast cameras, custom lighting rigs, a specialist AI company, and weeks of production. The deepfake that actually threatens your money, your perception of reality, and your safety costs about $40 in monthly subscriptions and two days of tinkering.

That shift — from expensive and impressive to cheap and good enough — is the defining feature of the AI deepfake threat in 2025 and beyond. Research from UC Berkeley shows that humans perform barely above chance when asked to distinguish real voices from AI-generated ones. For images, the result is the same. For video, accuracy sits around 70% — and that's under controlled lab conditions, without emotional manipulation, partisan framing, or time pressure. Professor Hany Farid, who runs perceptual studies at Berkeley, has said publicly that within 12 months, video deepfakes will be effectively indistinguishable to the average viewer.

That's not a dystopian forecast. It's a research finding. And the financial, political, and personal implications are significant.

---

How AI Deepfake Scams Actually Work

The mechanics of deepfake scams follow a straightforward formula: borrow a trusted identity, manufacture urgency, and extract money before the target has time to verify.

Celebrity impersonation is the most visible form. MrBeast, Elon Musk, Joe Rogan, and Donald Trump have all been deepfaked into fake product endorsements, cryptocurrency giveaways, and investment schemes. The playbook is consistent:

• A deepfaked video of a famous face endorses a product or financial opportunity
• Viewers trust the face, not the claim
• A QR code, link, or phone number directs them to a scam
• Money or personal data is extracted

One of the most persistent formats is the fake live stream — typically featuring a deepfaked US president or tech CEO running a "double your Bitcoin" scheme. YouTube has acknowledged these are getting harder to detect, and despite years of moderation efforts, they continue to resurface under new accounts.

But the more sophisticated development is the use of AI to replace human labour in scam operations entirely. Cybersecurity firm Checkpoint Research recently uncovered a piece of malware called VoidLink — described as "highly modular" and "resembling the efforts of multiple professional teams." Days later, they confirmed the code had been written almost entirely by AI, likely by a single operator, in under a week.

The implication is stark: scam operations that previously required a call centre, a management layer, and multiple human accomplices — all of whom could be arrested or turn informant — can now be run solo. Fewer people means lower operational risk and higher margins. AI callers don't sleep, don't demand raises, and critically, don't cooperate with investigators.

The key takeaway: the economics of fraud have fundamentally changed. Sophisticated, scalable scams are no longer the exclusive territory of organised crime groups.

---

The E-Commerce Deepfake Problem Nobody Is Talking About

Beyond celebrity impersonation, a quieter deepfake fraud is spreading through e-commerce. AI-generated "artisan" personas — typically elderly craftspeople portrayed in warm, handmade-looking product photos — are being used to sell cheap, mass-produced goods that look nothing like the advertised items.

The fake couple knitting sweaters by the fire. The elderly woodworker with the rustic workshop. None of them exist. They're AI-generated trust signals designed to exploit the premium consumers are willing to pay for authenticity and craftsmanship.

This represents a direct financial harm to consumers and a systemic trust problem for legitimate small producers who compete in the same market. When buyers get burned by fake artisans, they become sceptical of real ones.

---

Deepfake Propaganda: It's Not About Making You Believe. It's About Making You Tired.

Most people assume propaganda is something that happens to other people — less educated, more credulous, more politically extreme. That assumption is precisely what makes modern deepfake propaganda effective.

A clear example emerged during the US operation targeting Venezuelan president Nicolás Maduro. A wave of videos showing Venezuelans crying with joy — apparently celebrating the raid — went viral across social media. The videos were AI-generated. But by the time debunks circulated, the narrative had already shaped opinion in both directions: some people believed the videos uncritically; others, who saw them exposed as fake, became more suspicious of all coverage of the event.

That second effect is arguably more damaging and far less discussed. When propaganda is exposed, it doesn't just lose its power — it poisons the well. Verified, accurate information becomes harder to trust because it sits in the same media environment as the fakes.

Professor Farid puts it plainly: "The real danger is that we cognitively check out. We give up. It's too complicated. We don't know what to believe." That outcome — mass disengagement from factual inquiry — is not a side effect of propaganda. For authoritarian actors, it's the goal.

This dynamic is well understood in countries with longer exposure to state propaganda. The Russian perspective, captured in commentary from those who grew up under Soviet-era information control, is instructive: real propaganda doesn't arrive as a dramatic event. It doesn't shout. It talks until you're too tired to care.

Deepfake technology has made that process dramatically cheaper and faster to deploy. During the Russia-Ukraine conflict and the Israel-Gaza war, AI-generated fake videos and speeches circulated at scale. The effect wasn't uniform belief in any single narrative — it was a generalised erosion of confidence in video evidence as a category.

The practical risk for investors and business professionals: in an environment where video evidence of corporate misconduct, regulatory violations, or executive statements can be dismissed as potential deepfakes, accountability mechanisms weaken. Market-moving information becomes harder to verify. The cost of due diligence rises.

---

Platform Detection Is Necessary but Not Sufficient

YouTube is piloting a "likeness detection tool" that allows creators to flag unauthorised use of their image and voice. Dr. Mike Varshavski, a board-certified physician and medical content creator, is among the beta participants — and reports being deepfaked multiple times daily across various platforms.

The structural problem is clear: a detection system on one platform does nothing to prevent abuse on another. As Dr. Mike noted, bad actors simply migrate to the platform with the least enforcement. You are only as protected as the weakest platform in the ecosystem.

This creates a regulatory gap that individual platforms cannot close unilaterally. Some jurisdictions are beginning to legislate — several US states have passed laws targeting deepfake pornography and election-related deepfakes specifically — but enforcement lags significantly behind the technology's deployment speed.

For individuals, the honest answer is that platform-level tools provide partial protection at best. The burden of verification increasingly falls on the viewer.

---

How to Protect Yourself: A Practical Framework

Given that human perception is unreliable and platform tools are incomplete, the most effective defence is procedural rather than perceptual. Don't try to spot the fake. Instead, verify the claim through independent channels.

For financial decisions:

• Never act on investment information from a video alone, regardless of who appears to be speaking
• Any unsolicited offer involving cryptocurrency, especially "send X, receive 2X" formats, is a scam by definition — no legitimate financial product works this way
• Verify celebrity or executive endorsements by checking the official website, verified social accounts, or press releases directly
• Be particularly sceptical of live streams featuring prominent figures promoting financial products

For news and information:

• Treat viral videos of emotionally charged events — protests, arrests, military operations — with elevated scepticism until corroborated by multiple independent sources
• Look for the original source of the video, not just the account sharing it
• Blurry, imperfect footage is not evidence of fakery — in fact, hyper-crisp imagery in chaotic real-world situations can be a red flag
• Acknowledge that being uncertain is a valid and honest position. Suspending judgment is not the same as disengagement

For businesses:

• Organisations handling sensitive communications should establish verification protocols for any video-based authorisation, especially wire transfers or access credentials
• Assume that AI-generated voice clones of executives are technically feasible and factor that into security procedures

---

The Bottom Line

The deepfake threat is not primarily about a future where we can't tell anything apart. It's about a present where cheap, imperfect fakes are already good enough to defraud consumers, pollute political discourse, and erode institutional trust — at a scale that was previously impossible.

The Berkeley research data tells the story clearly: humans cannot reliably detect fake audio or images. Video discrimination sits at roughly 70% accuracy in ideal lab conditions. Within a short timeframe, that gap is expected to close further.

The response to this environment isn't panic or paralysis. It's building better habits around verification, maintaining healthy scepticism without sliding into cynicism, and understanding that the goal of much deepfake content isn't necessarily to make you believe something false — it's to make you stop trying to find out what's true.

That distinction matters, because the defence against the first is fact-checking. The defence against the second is refusing to give up on the process.

---

Frequently Asked Questions

What is a deepfake and how is it made?

A deepfake is a piece of media — video, audio, or image — in which a person is made to appear to say or do something they never did, using deep learning AI models. Early deepfakes required specialist equipment, large training datasets, and professional teams. Current consumer tools allow credible deepfakes to be created with a few dollars in software subscriptions and no technical background.

Can deepfake scams actually steal money from informed people?

Yes. The scams do not rely solely on the deepfake being undetectable. They rely on borrowed trust, urgency, and the fact that most people do not independently verify claims made by a familiar face. Research shows humans perform at near-chance levels when identifying fake audio and images, meaning even attentive viewers are vulnerable. The most effective defence is procedural: verify claims through official sources rather than trying to spot the fake visually.

How are deepfakes used in financial fraud specifically?

The most common formats include: fake celebrity endorsements of investment products or supplements; live streams of deepfaked executives or politicians running cryptocurrency doubling scams; AI-generated artisan personas used to fraudulently sell counterfeit or low-quality goods; and AI voice clones used in vishing (voice phishing) calls targeting individuals or businesses. Cybersecurity researchers have also documented AI-authored malware enabling sophisticated fraud operations run by single individuals.

What should I do if I suspect a video or audio clip is a deepfake?

Don't try to detect it visually — that approach is unreliable. Instead: search for the original source of the clip; check whether the claim in the video is corroborated by official statements or established news outlets; be especially sceptical if the content is asking you to act quickly or involves money; and treat emotionally charged or conveniently timed viral content as requiring extra verification. When in doubt, suspension of judgment is a reasonable and rational response.

Are there laws against deepfakes?

Legislation varies significantly by jurisdiction. Several US states have enacted laws specifically targeting non-consensual deepfake pornography and election-interference deepfakes. Federal legislation in the US remains fragmented. The EU's AI Act includes provisions relevant to synthetic media labelling. Enforcement consistently lags behind technological development, meaning legal protection is currently incomplete for most individuals and businesses.

---

This article is for informational purposes only and does not constitute financial advice. Always consult a qualified financial professional before making investment decisions.