World-Model Developers Stay Quiet on What They Are Building

World models — AI systems designed to automate spatial intelligence — have become one of the most closely watched and least transparent corners of the artificial intelligence industry. The companies chasing the technology have accumulated large funding rounds and considerable buzz, yet few are willing to describe what they are actually developing.

The field's best-known players are AMI Labs, co-founded by Yann LeCun, and World Labs, led by Fei-Fei Li. Both have attracted significant investment, but both also rank low on efforts to generate revenue, according to reporting on the sector.

Michael Rabbat, a co-founder of AMI Labs and the company's vice president of world models, took part in a panel discussion on the topic at the All In conference in September 2026. Pressed on what his company is working on, Rabbat said the company would discuss its work when it is ready to do so. He noted that AMI Labs is less than a year old, which makes a low profile reasonable for now.

That reticence is not confined to founders. Suppliers of the data used to train these systems are similarly guarded about their work, leaving outsiders with few authoritative details about how the technology is being built.

World Labs' Marble is regarded as the most fully developed product in the category so far. Its demonstrations cover straightforward media creation, the construction of explorable environments for video games, and computer-generated visual effects.

At their core, world models aim to give machines a working understanding of physical space, a capability that could apply to robotics, interactive video and more advanced self-driving systems. Where exactly the technology will first be commercialized, however, remains unclear even to those watching the sector closely.