A startup specializing in conversational AI claims its newest model is now convincingly human enough to pass as a real person during brief video calls, raising fresh questions about how quickly the line between people and machines is blurring online.
The Claim
Tavus, the company behind the model dubbed Griffin, says that in an internal test nearly half of participants mistook its AI for a flesh-and-blood human during short video interactions. According to the company, 26 of 54 people who took part in a one-minute video call came away believing they had been speaking with another person rather than software.
The figure amounts to roughly 48 percent of testers being fooled, a striking result if it holds up under independent scrutiny. It is worth noting, however, that the data comes from Tavus itself, and has not been verified by outside researchers.
Nearly half the testers walked away convinced they'd just chatted with a real human.
The Caveats
There are important limits to the announcement. The test involved a single short call lasting only about a minute, a format that gives an AI far less room to slip up than a longer, more probing conversation would. Self-reported results from a company promoting its own product also warrant a healthy dose of skepticism.
Griffin is not yet available to retail customers, meaning the public can't currently put the claims to the test. Tavus has positioned the model as a step forward in building AI agents capable of natural, face-to-face interaction rather than a finished consumer product.
- The test was based on a one-minute video call.
- Roughly half of 54 participants believed the AI was human.
- The results are internal and unverified by third parties.
- Griffin is not yet offered to retail users.
The announcement lands amid growing debate over the ethics and risks of increasingly lifelike AI, including concerns about deepfakes, impersonation, and the potential for deception. As conversational models grow more sophisticated, questions about disclosure and consent are likely to intensify, especially if systems like Griffin eventually reach the broader market.