Updates·October 4, 2026, 19:21

Tavus Griffin: AI face fooled almost half in video calls

AI-generated and checked against the sources listed below.

The US company Tavus says its new model Griffin was mistaken for a real human by 26 out of 54 test subjects. The model is not yet open to customers.

AI-generated image

The US AI company Tavus has shown a new model called Griffin. According to the company, it was perceived as a real human by almost half of the participants in a test.

Tavus presented the model on October 1, 2026. The company calls it a "Human Interaction Model." It is an AI that sees, listens and responds in a live video call with a generated face, voice and body language.

The test

In a blind test, participants talked for one minute over video with a Griffin-based AI human. Of 54 participants, 26, or 48 percent, believed they had talked to a real person.

Tavus' previous system fooled only 1 out of 41 participants, around 2.4 percent. The company therefore believes that Griffin is the first model to have passed a kind of "Turing test" for video. A Turing test is a test of whether you can tell the difference between human and machine.

The figures come from Tavus itself. The test is small and only one minute long, so it does not say that AI can always fool us.

What's new

Griffin is so-called "full-duplex." That means both parties can speak and react at the same time, as people do when we interrupt each other, nod or talk over each other. Many earlier AI avatars wait until you have finished speaking.

Not for customers yet

Griffin Lite is a research preview for selected testers and not a product that Tavus' customers can buy. The company is holding it back because the high realism can also be used to deceive. Tavus says it is working on features that reveal that you are talking to an AI, and on further safety measures, before the model is rolled out.

What it means for you

In the long run, you may encounter AI people in customer service, training or recruitment who seem like real humans. That makes it more important that companies disclose when there is an AI at the other end. It is also a reminder that you cannot blindly trust that a face on a screen is a human.

What it means for you

You cannot buy or try Griffin yet, but the technology shows where video calls are heading. If you or your workplace does customer service, recruitment or training over video, it is a good idea to decide already now that you always disclose when a customer is talking to an AI. As an ordinary user, you can be a little extra alert if a video call seems strangely smooth, and you can ask directly whether you are talking to a human.

Sources

More on this topic

Get the week's AI news in your inbox

Choose your level, topics and length. One email a week, unsubscribe at any time.

Subscribe to Promptly Newsletter
PromptlyNewsletterRSSLog in

The news on aijour is AI-generated and checked against the cited sources.