AI Video Model 'Griffin' Deceives Nearly Half of Call Participants

AI Mimicry Advances

Tavus, a company specializing in AI-driven video technology, announced that its latest model, Griffin, achieved a significant milestone in realistic human simulation. In a controlled test, 26 of 54 participants believed they were engaged in a video call with a real person, not an artificial intelligence.

Test Parameters and Current Status

The test involved one-minute video interactions. Tavus conducted the experiment internally, and Griffin is not yet available for commercial use. The company did not disclose details about the specific parameters of the AI’s responses or the demographic makeup of the test group. This level of deception highlights the rapid progress in generative AI capable of replicating human behavior in real-time video.

Implications for Deepfakes and Security

While Tavus focuses on applications like personalized video creation and sales enablement, the technology underscores growing concerns around deepfakes and the potential for malicious use. The ability to convincingly impersonate individuals via video raises questions about authentication protocols and the need for robust detection methods. The company has not yet detailed how it plans to mitigate potential misuse of the Griffin model.

Source: Decrypt