The Strange, Fast History of Deepfakes: From a Lab Trick to a $25 Million Heist

A finance worker in Hong Kong sat down for a video call with his boss. The chief financial officer was right there on screen. So were several other colleagues he recognized. They talked. They told him to send some money. He sent it — fifteen wire transfers, $25 million in a single day.

Every single person on that call was fake.

Not one of them was real. The faces, the voices, the little nods — all of it was generated by a computer. The worker only realized something was wrong days later, when he checked with head office and learned the meeting never happened (CNN, 2024).

How did we get here? How did a machine learn to wear a human face? The answer is a short, strange, and surprisingly fast story.

The Documented Facts

The word "deepfake" feels new, but the idea is older than you'd guess. Back in 1997, three researchers built a program called Video Rewrite. It took old footage of a person talking and re-stitched their mouth so they appeared to say brand-new words — words they never actually spoke. It was meant to help dub movies into other languages. It was clunky, but it was the first system to fully automate this kind of facial trickery (History of Information).

For years, the technology crawled. Then came a leap.

In June 2014, an AI researcher named Ian Goodfellow and his colleagues introduced something called a Generative Adversarial Network, or GAN (Wikipedia). Here's the clever part, and it's worth slowing down for. A GAN is really two computer programs locked in a contest. One — the "generator" — tries to make fake images. The other — the "discriminator" — tries to catch the fakes. Every time the discriminator wins, the generator learns and tries harder. They battle, over and over, thousands of times. The forger gets better because the detective keeps catching it (MIT Sloan). Out the other end comes something startlingly real.

The actual word "deepfake" arrived in November 2017 — and it came from the worst possible place. A Reddit user calling themselves "deepfakes" started posting fake videos that swapped celebrities' faces into pornography. The name stuck: "deep learning" plus "fake." A whole subreddit grew up around it, reaching almost 90,000 members before Reddit banned it in February 2018 (Reality Defender; The Verge).

Banning it did not make it go away. The tools were already out in the open.

Then the technology jumped from the internet's basement into the bank vault. In 2019, criminals used AI to clone the voice of a chief executive — the slight German accent, the melody of his speech, all of it — and phoned an employee at a UK energy firm. The worker, certain he was talking to his boss, transferred €220,000 (about $243,000) within the hour (Trend Micro). Five years later came the $25 million Hong Kong video heist. Each step, the fakes got cheaper, faster, and harder to spot.

The Genuine Open Question

Here's the part nobody has solved: can we reliably tell real from fake — and will we ever be able to?

You'd think we could just build a "fake detector." We tried. The trouble is that GANs were designed to beat detectors. Remember the two dueling programs? Teaching a machine to catch deepfakes just teaches the deepfake-makers what to fix next. It's a cat-and-mouse game where the mouse keeps getting smarter (Optica).

And the scale is exploding. By late 2024, large national banks were reporting more than five deepfake attacks per day, up from fewer than two per day at the start of that year (SecurityWeek). Researchers studying the problem in 2024 admitted that, despite real progress, detection still stumbles badly when it meets a new kind of fake it has never seen before (arXiv, 2024).

So the honest answer is: we don't have a dependable, future-proof way to spot a deepfake. The forgers, for now, are winning.

Theories and Interpretations

So where does this go? Here are the leading ideas — and to be clear, these are interpretations and predictions, not settled fact.

Theory 1: Detection eventually catches up (cautious optimism). Some experts believe better AI detectors, combined with "watermarking" — invisible tags baked into real videos at the moment they're filmed — could tip the balance back. Speculative: it depends on whether every camera maker actually adopts the same system, which is far from guaranteed.

Theory 2: We give up on detection and verify the source instead. Others argue we'll stop asking "is this video fake?" and start asking "can this video prove where it came from?" Think of it like a tamper-proof seal on a medicine bottle. Still unproven — the technology exists but is nowhere near universal.

Theory 3: The real danger isn't the fakes — it's us. One of the more chilling takes from researchers: the threat "comes not from the technology used to create it, but from people's natural inclination to believe what they see" (UNESCO). The flip side of a world full of fakes is the "liar's dividend" — when anyone caught on camera can simply shrug and say "that's a deepfake." This is an argument, not a measured fact, but it's a worrying one.

And the wilder claims? You may see online posts insisting that famous public events, moon landings, or world leaders are all secretly deepfakes, or that some shadowy group has been replacing real people for years. There is no credible evidence for any of that. Treat those as internet folklore — unverified, and almost certainly false.

Sources & Further Reading

• CNN — Finance worker pays out $25 million after video call with deepfake CFO (2024)
• History of Information — Video Rewrite, Origins of Deepfakes (1997)
• Wikipedia — Generative Adversarial Network
• MIT Sloan — Deepfakes, explained
• Reality Defender — A Brief History of Deepfakes
• The Verge — Reddit bans deepfakes face-swap porn community (2018)
• Trend Micro — Unusual CEO Fraud via Deepfake Audio Steals US$243,000 (2019)
• SecurityWeek — The AI Arms Race: Deepfake Generation vs. Detection
• Optica — Generating and Detecting Deepfakes: A 21st-Century Arms Race (2025)
• arXiv — Robust Deepfake Detection Requires More than Visual Forensics (2024)
• UNESCO — Deepfakes and the crisis of knowing

The deepfake started as a duel between two machines — one lying, one trying to catch the lie. But that same idea, two AIs locked in a contest, is now teaching computers something far stranger than how to fake a face. Some researchers think it's teaching them to want things. What happens when an AI stops imitating us and starts deciding for itself?