If you thought you knew what deepfakes were capable of, prepare for the plot to thicken. An AI startup called Tavus just dropped their new model, Griffin, and the numbers are honestly tripping people out. In a recent test, Tavus claims that 48% of people chatting with their 'Griffin-Lite' AI on a one-minute live video call walked away thinking they were talking to a real human.
The Breakdown
To get these stats, Tavus told participants they were being matched with another person for a quick chat about their plans for the year. No one was warned a bot might be on the other end. While the older version of their tech only tricked one out of 41 people, the new model had 26 out of 54 people totally convinced. Tavus is calling Griffin the first 'Human Interaction Model' because it pays attention to your expressions, pauses, and tone—not just the text.
But wait—don't ape into the hype just yet. The results come directly from the company's own research page. X (formerly Twitter) users have already pointed out that these findings aren't independently verified and don't follow standard academic protocols. Even Tavus admitted that those who got suspicious usually clocked the AI within 20 seconds.
Is It Actually Ready?
Real talk: Griffin-Lite is currently only available to a small group of 'trusted testers'—you can't just go sign up and start using it. When we tried to get under the hood, the results were, let’s be real, disappointing. Tavus says they are still building in safety features before a public release because they know how dangerous this tech can be.
Why it matters
We’ve already seen hackers use deepfakes on Zoom to impersonate coworkers and plant malware. If an AI can hold a full-duplex conversation—meaning it listens, watches, and talks simultaneously without that awkward 'walkie-talkie' delay—the bar for identity verification just got a whole lot higher. Security pros like those at Kraken have already been forced to use 'spontaneous questions' to catch imposters, but as AI gets smoother, staying safe means being extra cautious about who you’re actually talking to on camera. This is definitely not financial (or security) advice, but maybe start double-checking your sources before you trust a screen. Stay alert out there.
