I Created a Live Deepfake That Fooled 4 Out of 5 Friends at UC Berkeley

I created a live deepfake that fooled 4 out of 5 of my friends at University of California, Berkeley.

What you're seeing isn't me. It's an AI deepfake that looks exactly like TikTok's CEO, Shou Zi Chew, generated live from a single image.

If that sounds terrifying, it should.

The Reality of Real-Time Deepfakes

With the rise of tools like Sora from OpenAI and real-time AI video generation, the line between real and fake has nearly disappeared. After 7 years in cybersecurity, I've learned one thing: being proactive beats being reactive every single time.

Why This Matters for Your Organization

We've reached a point where seeing is no longer believing. Traditional security awareness training focuses on email phishing, but the threat landscape has expanded dramatically:

• Real-time video deepfakes can be generated from a single image
• Voice cloning can replicate anyone with just seconds of audio
• Social engineering attacks are becoming increasingly sophisticated

Building Veritas: Proactive Defense

That's why I'm building Veritas, a deepfake & voice-clone simulation platform that helps companies train their teams before scammers strike.

The goal is simple: prepare your workforce to identify and respond to AI-driven attacks through realistic simulations in a controlled environment.

Key Takeaways for Security Leaders

• Visual verification is no longer sufficient - Seeing someone on video doesn't mean they're real
• Multi-factor authentication is critical - Verify requests through multiple independent channels
• Employee training must evolve - Include deepfake and voice cloning scenarios in security awareness programs
• Test before attackers do - Simulate attacks to identify vulnerabilities in your human defenses

P.S. I promise the deepfake didn't fool them just because Mr. Shou Zi Chew and I are both Singaporean 😄