How to Detect Deepfakes: People Miss 75% of Videos

Imagine scrolling Instagram at midnight and seeing your country's central bank governor, in a video, telling you exactly which investment to put your savings into. He sounds right. He looks right. He's standing somewhere that looks official. Would you know, in that moment, how to detect deepfakes before you clicked "learn more"? Because that's precisely the trap Malaysia's central bank says criminals just set — using the actual face and voice of its own governor.

Bank Negara Malaysia has warned the public that scammers are running deepfake video ads impersonating its governor to push fake investment schemes, and the uncomfortable truth is that ordinary people are now expected to fact-check the very institutions they're supposed to trust.

What happened with the Bank Negara deepfake ads

Here's the part that should make your stomach drop: this isn't the first time. Back in May, fabricated videos circulated showing Bank Negara's governor supposedly caught in a physical altercation with a guest — overturned furniture, ambulance claims, the works. None of it happened. The governor had to appear on camera just to prove he was, in fact, not brawling with anyone. Now the same institution is warning about a different flavor of fake: deepfake video ads using his likeness to endorse investment products that don't exist, or that exist only long enough to take your money and disappear.

Think about what that means. The one authority figure Malaysians are trained to trust with financial guidance — the actual head of the central bank — has become the most useful mask a scammer can wear. That's not a coincidence. It's strategy. Criminals don't fake a random stranger's face. They fake the face that comes pre-loaded with credibility, because authority bias (our brain's shortcut of "this person is official, so this must be legit") does half the persuading for them before you've even read the pitch. Every deepfake relies on that same shortcut, which is exactly why identifying deepfakes early matters so much.

How to detect deepfakes when regular people are misled

Let's zoom out, because Malaysia isn't an isolated case — it's a preview of how deepfake technology is reshaping fraud everywhere. Globally, reported deepfake-fraud losses have hit at least $3.7 billion, and roughly 89% of that happened in just the last 18 months. That's not a slow creep. That's a hockey stick. And the single biggest category isn't romance scams or fake job offers — it's investment fraud, sitting at $900 million, or 57% of all deepfake losses tracked, according to Veriff's Identity Fraud Report.

Why investment fraud specifically? Because it's the perfect crime for a deepfake video. Nobody expects your bank governor to text you directly. But everyone accepts that officials give speeches, interviews, TV appearances. A fabricated version of that just... slots right into the format you already believe. And here's the technical detail that should really unsettle you: cloning someone's voice convincingly now takes as little as 20 to 30 seconds of audio, according to Fourthline's research on deepfakes in banking. Every press conference, every parliamentary hearing, every TV interview a public official gives is now raw material for someone else's scam script. This article is part of a series — start with Deepfake Audio.

Deepfakes do not simply impersonate identities, but elevate authority and familiarity to a convincing level. — Atlas Institute for International Affairs, Atlas Institute analysis

How to spot deepfakes and identify deepfake videos in an ad

Look for pacing that sounds a little too even, mouth movements that lag half a beat behind the audio, and lighting on the face that doesn't quite match the background. Blinking that's too rare or too regular is another tell, and skin can appear too smooth in ways real footage rarely shows. Pay attention, too, to obvious borders around the face where the digital overlay meets the real background — a common sign in unsuccessful screen captures of a deepfake video. None of these guarantee a fake — but stacked together, they're a signal to stop before you click, share, or send money.

---

Deepfake detection tools versus human judgment

Now, the tech industry loves to say deepfake detection tools will save us. Gartner analysts have predicted that by 2026, 30% of companies will lose confidence in facial biometric authentication specifically because AI-made fakes have gotten that good. That's a company-side crisis of confidence — imagine what it does to a regular person with no fraud team and no forensics lab.

Because here's the kicker: in controlled studies, people spotted deepfake videos correctly only 24.5% of the time. Flip that number around and it says something brutal — three out of four fakes slipped past human eyes completely undetected. So when someone tells you to just "look closely" at a suspicious video, that advice is, frankly, not good enough on its own. This is exactly the gap deepfake detectors are supposed to fill, and it's also why some researchers built dedicated deepfake detection projects that score suspicious clips automatically. One such project, the Deepfake-O-Meter's algorithm, distinguishes successful fakes from more obviously flawed ones by scanning for the same mechanical inconsistencies a trained eye would look for, just faster and at scale.

What you can actually do before you click "invest"

Okay, so if you've ever stared at an ad and thought, wait, would that person really say this? — that instinct is exactly the right one, and it's worth trusting more, not less. Here's the one concrete thing to do: before you act on any video claiming to show a bank official, government figure, or financial authority recommending a product, go directly to that institution's official website or verified social account and search for the same announcement there. Not a link in the ad. Not a comment section. The actual homepage, typed in yourself. If Bank Negara really endorsed something, it would be sitting on bnm.gov.my in plain text, not buried in a Facebook ad with a "limited time" countdown clock.

That single habit — verify at the source, not at the link — would have stopped both the fake fight video and the fake investment ads cold, because neither one exists anywhere except in the scam itself. It also beats relying purely on any single detection tool, since no deepfake detector, however good, catches every fake video every time.

The bigger deepfake problem authorities haven't solved

Malaysia's own cybersecurity data backs this up at a national scale: MIMOS Berhad has tied RM1.58 billion in 2024 cybercrime losses directly to deepfake-driven fraud. That's not a rounding error in a national budget. That's an entire industry of fake authority, running quietly in the background of every ad platform that hasn't figured out how to screen for synthetic video before it goes live.

And this is where I'll say the quiet part out loud: the current setup asks you, a person with a phone and maybe two minutes of attention before your coffee gets cold, to out-detect fraud that fools trained analysts a majority of the time. That's not a fair fight. Ad platforms accepted money to run these deepfake video ads. Somewhere in that transaction, nobody checked whether the central bank governor actually said any of it. Fixing that isn't a "media literacy" problem for citizens to solve on their own time. It's a gatekeeping failure, and the gate is the one that should have been locked in the first place.

Deepfake fraud protection question people are asking

People keep asking whether banks will start adding some kind of verification stamp to real videos, or whether better deepfake detection tools will simply solve this for us. Right now, no major institution has rolled that out publicly — which means, for now, the burden of checking still falls on whoever's watching the ad at 11pm, wallet in hand.

So next time a familiar face tells you where to put your money, ask yourself the blunt version of the question: if the actual governor of a central bank can't stop people from faking his own fistfight, what makes any of us think we'd catch the quieter, better-produced deepfake selling us an "investment opportunity" instead?

Frequently Asked Questions