Why AI labs are keeping their new projects a secret

The newest focus for AI labs is the development of world models. Unlike a standard chatbot that just predicts the next word in a sentence, a world model is designed to understand spatial intelligence—the ability to perceive and interact with the physical world. While researchers are building these systems to handle tasks like guiding robots or creating explorable video environments, they are refusing to share what they are actually working on. Even companies providing the raw data for these models don't know the end goal. Behind the scenes, some AI CEOs are calling for new safety protocols and outside observers to monitor their work, while others, most notably the CEO of chipmaker Nvidia, are arguing that these fears are unnecessary and that regulation would only hold back progress.

Think of a standard AI model like a digital librarian who has read every book in the world but has never walked outside. It knows the dictionary definition of a bridge or a car, but it doesn't intuitively understand how gravity, physics, or three-dimensional space work. A world model acts more like a simulation or a map of the physical environment. By ingesting massive amounts of video data, the model learns the rules of the world—how objects fall, how light moves, and how spaces fit together. This is the same logic used to help self-driving cars navigate traffic. Once a model has this spatial awareness, it can be applied to almost anything: it could allow a robotic arm to pick up fragile items, create realistic video game landscapes from a few camera clips, or even help doctors navigate complex medical imagery.

The secrecy around these projects highlights the strange economics of the current AI boom. Because money is easy to raise, companies are hoarding their ideas to stay ahead. They are operating in what some call a dark forest: in a crowded market where any breakthrough can be quickly copied, the safest move is to remain invisible until a product is ready to launch. Meanwhile, the public debate over slowing down is less about stopping AI and more about who gets to hold the steering wheel. When executives talk about pacing the industry, they aren't necessarily calling for a halt; they are proposing a system of safety checks that could eventually shape the rules for everyone. Whether these companies actually intend to slow down, or are simply positioning themselves for the next phase of competition, remains an open question.