When a Deepfake Promises Rp 75 Juta, the Scammers Win Before You Even Click

I keep coming back to this one because it reveals something about how AI deception works now. The Gibran video wasn't some clever hoax designed to make you laugh or rethink your politics. It was engineered to make you part with money.

A video of Vice President Gibran Rakabuming Raka promising to distribute Rp 75 million in subsidy funds circulated on social media. The narrative was simple: apply through this Messenger link to receive your allocation. Sound familiar? You've seen that pitch before, from real politicians and from scammers who don't bother deepfaking anyone.

The twist is the forensic evidence that followed. Kompas's fact-check team ran the video through Hive Moderation and TruthScan. The results: 99.9% probability the video was AI-generated, 98.5% probability the audio was synthetic source. The West Java police public relations office also flagged it source. And the link? It redirected to a Facebook Messenger account called "Program dana bantuan Gibran," which was almost certainly a phishing funnel.

Here's what I find interesting about the mechanics. The deepfake quality is high enough to fool casual viewers scrolling through their feed. But it's low enough that forensic tools catch it immediately when they run the analysis. That gap -- between what the eye believes and what the algorithm detects -- is where the whole ecosystem lives right now.

I'm genuinely curious about what's happening under the hood. Most AI detection models are trained on known generators: Stable Diffusion, Midjourney, Runway, Pika. But as generators evolve, the models that trained to catch them will lag. This isn't a cat-and-mouse game with a finish line. It's an arms race that shifts the ground under everyone's feet every few months.

The more specific takeaway, though, is about the scam itself. The deepfake is the hook. The payload is the link. And the link leads to a platform -- Messenger -- where the conversation is private and the paper trail is minimal. You hand over your data, they take your money, and Facebook's moderation tools probably never see what happened inside that chat.

I keep thinking about the people who saw this video and didn't know whether to believe it. For someone who doesn't follow AI forensics, the video looks real. The voice sounds like Gibran. The face moves like him. The offer sounds like something a politician might actually propose. Why would you doubt it?

The answer is: you shouldn't trust what you see, especially when money is involved. But the deeper problem is that the threshold for "looks real" is dropping faster than our collective ability to spot the fake.

Sources: Kompas Cek Fakta | Polda Jabar