Zoom announced a new feature today aimed at transforming users into AI-driven, photorealistic avatars, with a release planned for 2025. This feature, unveiled at Zoom’s annual developer conference, will enable users to create digital clones of themselves by recording a short video clip. This AI-generated avatar will mimic the user’s head, shoulders, and upper arms, and users can type scripts for it to “speak,” with Zoom syncing the audio to match lip movements. This development is part of Zoom’s effort to facilitate “asynchronous” communication for faster, more efficient interactions, according to Smita Hashim, Zoom’s Chief Product Officer.
Zoom’s custom avatar feature aims to save users time by simplifying video creation, but it also raises ethical and security concerns around potential misuse. In recent years, several companies, like Tavus and Microsoft, have released similar technology, allowing brands and users to create virtual doubles that can convincingly speak using synthetic voices. These companies have implemented strict safeguards to prevent misuse, such as Tavus’ verbal consent requirements and Microsoft’s requirement for written permissions. However, Zoom has only briefly addressed its planned security measures.
In her statement, Hashim mentioned that Zoom plans to include “numerous safeguards” such as “advanced authentication” and watermarking to distinguish AI-generated videos. The company’s approach, however, remains somewhat vague, relying on Zoom’s usage policies to deter potential misuse. According to Hashim, Zoom’s watermark will signal avatar-generated content, and the company will implement further measures to maintain content authenticity. But given the ease with which watermarks can be cropped or removed, questions remain about how effective these safeguards will be.
Zoom’s AI-generated avatars are part of CEO Eric Yuan’s broader vision to make AI a significant component of Zoom’s platform. The long-term goal is to develop AIs that could manage basic tasks for users, like attending meetings, responding to emails, and taking calls. However, Zoom’s move into AI avatar technology comes amid a surge in deepfake content, which continues to spread misinformation and blur the line between real and fabricated media.
High-profile deepfakes of public figures, including President Joe Biden and Taylor Swift, have amassed millions of views this year. Other AI-generated images and videos, including fabricated footage following Hurricane Helene, have contributed to this trend, heightening concerns around the ethical use of generative AI. Deepfakes have also been exploited in impersonation scams, which the FTC reports led to over $1 billion in losses last year alone.
Given the risks, the lack of clear security measures in Zoom’s feature has raised questions about how the company plans to address the potential for AI-driven misinformation and scams. Although Zoom’s mock-ups display a watermark to identify avatar content, the effectiveness of this measure is uncertain since watermarks can often be removed with simple editing tools.
Regulators are beginning to take action to address the growing threat of deepfakes. While there is currently no federal law criminalizing deepfakes, more than ten states in the U.S. have introduced laws targeting AI-enabled impersonation. California, for instance, is considering legislation that would allow courts to order deepfake content be taken down and impose financial penalties. As Zoom moves toward releasing its avatar feature, the company will likely need to clarify its security measures further to align with these regulatory efforts and protect against potential misuse.
