💸 AI Helped Detect a Deepfake CEO (And Saved Millions)
A finance team almost wired $25 million to a fake CEO — until AI caught the scam.
Yes, you read that right. This is one of the craziest “AI saves the day” stories this year.
Here’s what happened:
A multinational company in Hong Kong got an urgent video call from their CEO.
He said the company had to move $25M to a new account for an “important secret deal.”
The video looked real. The voice sounded real. The background even showed his real office.
But here’s the catch — it wasn’t him.
It was a deepfake.
The scammers had trained an AI model using old interviews, public photos, and online clips of the CEO. They used generative video and voice tools to make a perfect copy — basically a human clone made of pixels.
The finance team almost approved the transfer… until a small AI-based fraud detection system flagged something odd — the video metadata didn’t match the usual meeting software logs.
They double-checked with the real CEO.
And boom! Total scam averted. $25 million saved.
Here’s what made this possible:
The AI system didn’t “see faces” like we do — it analyzed pixel patterns and found micro-glitches that humans miss.
It detected a 0.3-second delay between voice and lip movement — something no human eye noticed.
The system used previous real recordings of the CEO as a “voiceprint baseline” — and this one didn’t fully match.
So yes, AI just saved humans from another AI. Looks like AI said, “I made this mess, I’ll clean it up too.”
Why this matters:
This story isn’t just about cybersecurity, it’s about trust in the AI era.
As deepfakes get more real, companies are starting to use AI against AI.
If you’re learning web development, this shows how important verification, data logs, and pattern analysis are becoming in tech systems.
And if you’re a creator or brand owner, this matters for you too.
Soon, you might need to verify your own content, to prove you made it.
Kind of wild, right?
My takeaway
AI is getting smart, but also, so are the scams.
The best part? The same technology that can fake you can also protect you.
And that’s exactly the balance we’ll see more of.
Also, let’s be honest, if my boss suddenly called me asking for $25 million, I’d hang up. (Unless he promised me a Tesla.)
💡 Big Lesson:
You don’t need to be an AI expert to use it smartly.
You just need to know where to ask, “Wait, should I trust this?”
🧠 Prompt Tip of the Day
Ever notice how every ChatGPT reply sounds a bit... robotic?
Yeah, blame the em-dash (—) — the overused “AI signature” of our time. Once poetic, now just a red flag that screams “machine wrote this!”
But good news — Ruben Hassid dropped a genius guide on how to fix it. He explains how to make ChatGPT actually write like you.
Here’s the move:
Instead of saying “stop using em-dashes”, give ChatGPT a replacement rule.
Go to Settings → Personalization → Custom Instructions and add this line:
“Systematically replace em-dashes (‘—’) with a period (‘.’) to start a new sentence, or a comma (‘,’) to continue the sentence.”
Why it works:
AI needs structure, not scolding. Saying “don’t do this” doesn’t work — it’s like telling a toddler not to spill the juice (guess what happens next?).
Give it options instead.
Key insight:
If you want AI to sound like you, don’t hunt for a single magic prompt.
Train it bit by bit — teach it your writing habits, slang, and rhythm.
Basically, you’re giving AI your accent.
Thanks for reading!
Stay curious. Stay creating.
Shoikot Sazzad
✉️ From the newsletter Wait, I Should Know That
