Employees at Elon Musk's artificial intelligence company xAI recorded their own faces while they were talking, later using the videos to train Grok to interpret human emotions. More than 200 people participated in the internal project, many of whom were concerned about how their recordings might be used, Business Insider reported.
The project began in April 2025 and was called Skippy. Back then, AI teachers—xAI employees who help train Grok—recorded videos of themselves talking to their colleagues over the course of a week. In the 15- to 30-minute conversations, one person acted as a virtual assistant, or “host,” while the other took on the role of the user. The “host” had to minimize their movements to get the shot right, while the user acted completely naturally, simulating a casual conversation with a friend.
At the time, one of the lead engineers on the Skippy project was telling employees that the collected footage would help Grok “get a face.” The project manager noted that the data would later be used to create “human avatars.” By the way, xAI recently introduced animated avatars of Annie and Rudy, and while it’s not clear whether they used employee videos for them, it’s likely that they did.
As for the concerns of the employees involved in the project, management assured them that the facial recordings would not be released outside the company. One engineer also said that the employees' faces would never be used in production and that "this is done solely to train Grok."
Employees participating in the filming also had to sign a consent form that gave xAI “permanent” access to the data, including images of the employees, for training and “inclusion and promotion of commercial products and services offered by xAI.” The form specifically stated that the collected data would only be used for training purposes and not to “create a digital version.”