Tavus’ Griffin AI avatar fooled nearly half of test subjects in a one-minute video call

Tavus has introduced Griffin, which it calls the first “Human Interaction Model.” The system is designed to carry on face-to-face video conversations in real time while processing speech, facial expressions, tone of voice, gestures, and pauses. It also receives and generates video during the interaction, positioning it as more than a standard chatbot or prerecorded avatar.

In a Tavus study, 48 percent of participants believed Griffin was a real person after a one-minute video call. The article notes that previous systems maxed out at two percent, making the reported jump a major signal for AI video realism. Tavus also describes an independent Nvidia test in which Griffin scored 3.83 on how human an AI feels in direct audio-video conversation, compared with 3.92 for actual humans and 2.80 for the previous best AI model.

A preview called Griffin-Lite is available to select testers as a research preview. Tavus says a more capable version will come after safety concerns are addressed. The company lists tutoring, practicing difficult conversations, and camera-based tech support as potential use cases, all areas where real-time visual and verbal feedback could matter.

Griffin highlights how quickly AI video personas are moving toward live, responsive interaction. If systems can convincingly interpret nonverbal cues while speaking naturally on camera, users may need clearer signals about when they are interacting with AI. The immediate takeaway for teams testing this technology is to evaluate not only realism and usefulness, but also disclosure, safety, and trust boundaries.