Back to News
NewsAi

Tavus Unveils AI That Deceives Users During Video Calls

Tavus has launched its Griffin AI, achieving 48% human-like interaction in video calls, significantly surpassing its predecessor.

Mercation Admin
October 2, 2026 2 min read
Tavus Unveils AI That Deceives Users During Video Calls

Tavus Introduces Griffin AI with Enhanced Human Interaction Capabilities

Tavus has unveiled its latest artificial intelligence model, Griffin, which has demonstrated a significant improvement in simulating human-like interactions during video calls. According to the company's research, 48% of participants believed they were conversing with a real person during a one-minute demonstration with Griffin-Lite, a version of Griffin, while only 2.4% believed the previous model was human-like.

Impressive Benchmark Scores

Griffin-Lite achieved a score of 3.83 out of 5 on Nvidia's VideoFDB benchmark, outpacing other systems, including the next-best at 2.80 and a human reference at 3.92. This evaluation assesses the quality of live audio and video dialogue, indicating Griffin's superior performance in generating natural and expressive responses.

Research Methodology and Findings

The findings stem from proprietary research conducted by Tavus, where participants were misled into believing they were paired with another human for a discussion about their future expectations. After the interaction, only then were they asked if they suspected their partner might not be real. This approach deviates from the traditional Turing test, which involves a judge discerning between human and machine.

Despite the promising results, a community note on X has raised concerns about the lack of independent verification of Tavus' findings and adherence to standard testing protocols. Participants often grew suspicious within the first 20 seconds of interaction.

Technological Advancements and Security Implications

Griffin's capabilities extend to full-duplex communication, enabling simultaneous listening and responding, akin to a natural phone conversation. This technology becomes increasingly relevant in light of rising security concerns, as scammers have begun using video calls for impersonation. For instance, North Korean hackers have employed deepfake technology to manipulate victims into installing malware during video calls.

Organizations are already adapting their security protocols in response to these advancements. Tavus has indicated that Griffin requires further safety measures before it can be made widely available. Currently, access to Griffin-Lite is limited to select trusted testers as the company focuses on enhancing safety protocols and developing disclosure features.

In November 2025, Tavus secured a $40 million Series B funding round to support ongoing development. Those interested in participating as trusted testers can apply through the Tavus website.