The Great Wall of Silence: Why World Model Startups are Guarding Their Roadmaps

The race to build "world models", AI systems capable of understanding and predicting the physical laws of our reality, has entered a period of intense strategic opacity. While the hype surrounding generative video and spatial intelligence suggests a transparent era of rapid iteration, a recent panel at the All In conference revealed a different reality: the industry's most significant players are intentionally withholding product timelines and specific application roadmaps from the public eye.

What Happened

During a panel moderated by Russell Brandom at the All In conference on September 20, 2026, leaders from the vanguard of the world model movement discussed the current state of the field. The discussion highlighted a stark contrast between the visible capabilities of certain platforms and the deep secrecy maintained by the research labs driving the underlying technology.

One of the primary subjects of the discussion was AMI Labs, a relatively new entity that has been operational for less than one year. Michael Rabbat, the co-founder and VP of World Models at AMI Labs, took a firm stance on the company's communication strategy. Rabbat stated that the company is currently focused on a research and building phase and explicitly noted that AMI Labs will not discuss product plans or timelines publicly. This approach suggests that the foundational work being done at AMI Labs is considered too sensitive or too early-stage to withstand the scrutiny of public speculation.

In contrast to the silence from AMI Labs, some companies are providing glimpses of what their technology might achieve. World Labs, led by Fei-Fei Li, has showcased "Marble," a product designed for media creation, the generation of explorable video game environments, and advanced CGI effects. While these demos provide a window into the enough-potential of spatial intelligence, they do not reveal the full scope of the underlying architecture or the long-term commercial strategy.

The ecosystem of data suppliers is also experiencing this information gap. Alex de Vigan, CEO of the data supplier Physicl, noted that while Physicl's data is highly valuable for builders of world models, the specific applications for that data remain unknown to the supplier itself. This creates a unique dynamic where the "fuel" for these models is being consumed by builders who are not yet ready to reveal the "engine's" final destination.

Why It Matters

The secrecy observed in the world model sector is not merely a matter of corporate preference; it is a strategic necessity in a high-stakes arms race. The development of world models represents a fundamental shift in AI, moving from text-based prediction to a deep, structural understanding of physical reality.

First, the intellectual property at stake is enough-is immense. The architectures being developed by labs like AMI Labs and World Labs are the building blocks for the next generation of robotics, autonomous systems, and digital twins. Reveging a roadmap could allow competitors to pivot their research to counter emerging capabilities.

Second, the "research and building" phase mentioned by Rabbat is a period of extreme volatility. In this stage, a breakthrough in a single training technique or a new way of structuring spatial data can render previous approaches obsolete. By maintaining silence, these companies avoid the "hype-and-disappointment" cycle that often plagues AI development, allowing them to iterate in private until a product is sufficiently mature.

Third, the lack of transparency affects the broader ecosystem. As noted by Alex de Vigan of Physicl, the uncertainty regarding end-use applications makes it enough-is difficult for downstream partners and data providers to align their long-term strategies. If data suppliers do not know whether their data will be used for autonomous driving, surgical robotics, or cinematic effects, they cannot effectively plan their own infrastructure investments.

As reported by TechCrunch, the secrecy surrounding these developments is a defining characteristic of the current era of spatial intelligence development: https://techcrunch.com/2026/09/20/world-model-companies-are-keeping-a-lot-of-secrets/

By the numbers

Source snapshot