Meta Builds Photorealistic AI Replica of Mark Zuckerberg

Meta is building a photorealistic, AI-powered 3D replica of CEO Mark Zuckerberg to hold real conversations with employees and serve as an always-available form of leadership, according to people familiar with the project. The avatar is being trained on his mannerisms, vocal patterns, public statements and recent strategic views.

The system is fed recordings of his voice, video of his gestures and internal and public communications so it can replicate unscripted conversations, answer questions about company strategy and stand in for the CEO in one-on-one interactions. Engineers say the goal is to make employees feel more connected to the founder and to provide scalable access to leadership across a global workforce.

The effort is housed in a newly formed Superintelligence Labs team and has been tested and trained by Zuckerberg. Scaling the technology requires large amounts of computing capacity to keep interactions smooth and low-latency. The work combines recent advances in text, speech and 3D rendering and has required acquisitions and infrastructure investment.

Last year the company acquired two voice-technology firms, PlayAI and WaveForms, to speed speech synthesis capabilities. Company planning documents project capital expenditure for 2026 between $115 billion and $135 billion to support heavier AI workloads.

The company released Muse Spark last week, a compact model from Superintelligence Labs that the company says has capabilities in health reasoning and visual understanding. The announcement was followed by a 7% rise in the company’s share price.

Inside the company, staff are being encouraged to adopt AI tools and to build their own assistants using open-source software called OpenClaw. Product managers have been given a “skills baseline exercise” that includes system design tests and tasks labeled “vibe coding.”

The initiative follows earlier metaverse work. In 2022 the company promoted Horizon Worlds and used a blocky avatar for Zuckerberg that drew widespread criticism. Reality Labs recorded heavy losses, including a reported $10.2 billion loss in 2021, before leadership redirected resources toward generative AI and internal tooling.

People involved in the avatar project described the aim as twofold: increase direct access to senior leadership for employees and automate routine interactions that would otherwise require the CEO’s time. The replica is intended to be available for scheduled and impromptu conversations and to provide consistent messaging on company priorities and processes.

Engineers and product teams face privacy, accuracy and governance challenges. Creating a convincing digital leader requires realistic audio and visuals and safeguards to prevent the model from making inaccurate statements or giving unauthorized commitments. The company has not provided public details about consent, auditing or limits on the avatar’s use, and internal testing is ongoing.

The company declined to comment on specifics of the project. According to people familiar with the work, development is active but a full-scale rollout will depend on solving technical and policy challenges, including compute costs and internal controls for how the replica represents and conveys leadership decisions.