Tavus AI Fools 48% of Humans in First Video Turing Test Pass

Tavus's Griffin model, a Human Interaction Model, allows users to have face-to-face conversations with a synthetic agent that can see and hear simultaneously. In a company study, 48% of participants believed they were speaking with a real person, and the model scored 3.83 out of 5 on Nvidia's VideoFDB benchmark. Tavus is currently limiting access to a Griffin-Lite version for trusted testers while implementing safety and disclosure measures.