AI Called Netanyahu's Café Video a Deepfake. It Wasn't. That's the Real Problem. | Podcast

Full Episode Transcript

An A. I. chatbot looked at a real video of Benjamin Netanyahu sitting in a café — and declared it a hundred percent deepfake.

It wasn't. And what happened next broke something

It wasn't. And what happened next broke something fundamental about how we trust video evidence. This past March, Netanyahu released footage to counter rumors about his health.

Instead of settling the debate, Grok — the A. I. chatbot — flagged the video as fabricated.

That forced a bizarre chain of events. Netanyahu's team had to produce café photos, metadata, even face-matching — round after round of proof-of-life authentication. Each new piece of evidence just triggered more skepticism.

If a head of state can't prove a video of himself is real, what happens when your footage lands in a courtroom? Three independent expert teams analyzed that café video using multiple detection tools. According to researchers at GetReal Security — co-founded by U.

Trusted by Investigators Worldwide

Run Forensic-Grade Comparisons in Seconds

Full platform access for 7 days. Run real searches — no credit card, no commitment.

Run My First Search →

C. Berkeley professor Hany Farid — none of them

C. Berkeley professor Hany Farid — none of them found significant evidence of A. I.

manipulation. The video was authentic. But Grok didn't just get it wrong.

It dressed its false conclusion in fabricated citations and authoritative framing. A tool people treat as a neutral referee was actually generating confident-sounding misinformation. Now look at what's happening in courtrooms.

The Federal Rules of Evidence still rely on a standard from the twentieth century. A witness with personal knowledge testifies that a video fairly represents what happened. According to the American Bar Association, that bar is extremely low.

Meone familiar with a person's voice could

Someone familiar with a person's voice could authenticate a recording, and courts have said that's probably enough to get it admitted. Does that standard hold when anyone with a laptop can fabricate photorealistic footage? Criminal defendants are already claiming prosecution videos are deepfaked.

Civil litigants are submitting A. I. -generated content to support false claims.

And remember Netanyahu's vanishing ring? In one frame, his ring seemed to disappear. Almost certainly a compression artifact or frame-rate drop.

But online, people instantly diagnosed it as an A. I. rendering failure.

The Bottom Line

That's the new default — every glitch becomes evidence of fabrication. The biggest threat isn't that deepfakes are undetectable. It's that the tools we trust to detect them are generating false confidence in both directions — calling real videos fake and giving fake videos a pass.

So — a real video got labeled fake by an A. I. tool.

The person in the video couldn't prove it was real no matter how much evidence he produced. And the legal system still authenticates video the same way it did decades ago. Courts are starting to respond — proposing pretrial hearings for authenticity disputes, requiring expert testimony for deepfake allegations, and raising the burden for video in high-stakes cases.

Watch for amendments to Federal Rule nine-oh-one. The written version goes deeper — link's below.

AI Called Netanyahu's Café Video a Deepfake. It Wasn't. That's the Real Problem. | Podcast