Secretive world models: why AI builders should watch the hush around AMI Labs and World Labs

At a recent AI conference, Michael Rabbat, co-founder of AMI Labs, offered little when pressed on product plans: "We'll talk about it when we're ready to talk about it." That caginess is the norm in the world models space, where two heavily funded labs AMI Labs and World Labs are building automated spatial intelligence systems but disclosing almost nothing about what they're actually shipping. For AI builders evaluating dependencies or future integrations, this wall of secrecy matters.

Automated spatial intelligence with no public roadmap

World models aim to give AI a persistent understanding of physical space, enabling robotics, interactive video, and self-driving systems. AMI Labs, led by Yann LeCun and less than a year old, remains in a "research and building phase" with no disclosed product plans or timelines TechCrunch. World Labs, founded by Fei-Fei Li, has shown Marble, a platform that generates explorable environments for video games and CGI effects, plus the Nabia AI doctor software through a partnership. Yet even these demos feel more like capability showcases than commercial products.

The secrecy extends to data suppliers. Alex de Vigan, CEO of Physicl, told TechCrunch his company supplies data to world model firms but doesn't know exactly what they're building TechCrunch. "I wish they would tell us more. We could build more useful data if we knew what they were working on."

World models are versatile. The same approach that helps a Waymo navigate traffic can help a humanoid robot carry boxes or turn video footage into an explorable scene. This breadth is part of why companies stay quiet: announcing a focus on one application would invite rivals TechCrunch.

Why this secrecy matters for builders

If you're building a product that depends on world model capabilities such as agent navigation, spatial reasoning, or scene generation you can't rely on a public timeline from these labs. AMI has explored manufacturing, biomedicine, robotics, and AI software for doctors, but it's unclear which paths will become products TechCrunch. The risk of building on a secretive platform is that direction changes without warning.

The secrecy is strategic. If AMI or World Labs revealed a killer product, competitors like OpenAI or Anthropic would quickly enter the space. The easy fundraising climate means these labs can stay private longer, but it also funds potential rivals once the path to market becomes clear TechCrunch. For builders, that means any integration with these systems carries high uncertainty.

Practical steps for evaluating world model tech

First, prioritize partners with transparent roadmaps and open data practices. Physicl's CEO highlighted that even suppliers operate in the dark, so downstream customers should expect even less visibility TechCrunch. Second, treat world model capabilities as experimental. Build abstractions that allow swapping in alternative spatial AI solutions as the market matures. Third, watch for signals from larger AI labs entering the space, which could accelerate product clarity or increase competition.

Caveats to keep in mind

The evidence on product plans is almost entirely negative: the companies haven't disclosed them. The TechCrunch reporting confirms that both AMI Labs and World Labs are deliberately opaque, and even their own data suppliers are guessing. The field is in a research and building phase, so don't expect stable APIs or SLAs anytime soon.