This AI Is Already Fooling People on Video Calls Into Thinking Its Human, Company Says
Tavus says its new live video AI model, Griffin, made 48% of test participants think they were speaking with a real human on a one-minute video call. Its earlier system scored 2.4% on the same internal test. Griffin-Lite, the tested version, combines real-time understanding and generation of face-to-face conversation, including expressions, pauses, voice, and vision, and Tavus says it is full-duplex and low-latency. The results are based on Tavus’ own research page and are not independently verified; participants were not told a bot might be present. Griffin-Lite also ranked highly on NVIDIA’s VideoFDB benchmark, though it still trailed humans on perception. Tavus says the model is only available to trusted testers for now while it adds disclosure and safety features. The launch highlights growing concerns about AI-assisted scams and deception in live video calls.
