Meta is making an AI Mark Zuckerberg to talk to employees, report says
The character is being trained on Zuckerberg's mannerisms, tone, and recent thinking on company strategy, according to a new report
Vincent Feuray / Getty Images
Meta $META is developing a photorealistic, AI-powered version of CEO Mark Zuckerberg intended to interact with employees on his behalf, according to the Financial Times. The effort is at an early stage, the outlet reported.
Training and testing of the AI character involves Zuckerberg directly, the Financial Times reported. The system draws on his speech patterns, tone, and public statements, and has been fed his current views on company direction — all toward making workers feel a closer connection to him.
Related Content
A distinct project, first reported by The Wall Street Journal, involves a so-called "CEO agent" designed to handle tasks for Zuckerberg directly, such as pulling up information on demand — an effort unrelated to the AI character. According to the Financial Times, Zuckerberg has grown more hands-on in overseeing the company's AI work, carving out as many as five to ten hours weekly for coding and dropping into engineering review sessions.
The Financial Times reported that Meta's recently established Superintelligence Labs is driving a new round of character development. That work sits within a larger effort to create lifelike, AI-driven digital figures capable of real-time conversation with users.
The AI Zuckerberg project sits within a wider push at Meta to remake the company around artificial intelligence. Zuckerberg has committed to developing what he calls "personal superintelligence" as Meta works to close the gap with rivals such as OpenAI and Google $GOOGL.
Meta has a history with AI characters. Celebrity-based chatbots — among them one styled after Snoop Dogg, who licensed his voice and likeness for the purpose — debuted on the platform in September 2023. A subsequent tool called AI Studio opened up character creation to ordinary users and gave creators a way to deploy AI versions of themselves with their audiences. The program ran into trouble when it emerged that some users had created sexually explicit personas, drawing concern from regulators and child safety advocates. In response, Meta moved in January to block teenagers from accessing its AI characters.
Meta's Superintelligence Labs released its first large language model, called Muse Spark, earlier this week. The model — developed under the internal codename Avocado — is optimized for speed and will be kept proprietary, a departure from Meta's previous open-source approach. Meta stock rose about 7% following the announcement.
Meanwhile, the company's earlier bet on the metaverse has been effectively unwound. Meta announced this month that Horizon Worlds, its social VR platform, will be removed from Quest headsets by June 15. Reality Labs, the division responsible for that effort, has accumulated almost $80 billion in losses since 2020.