The companies building the most important AI of this decade won’t tell you what they have. World models sit at the center of a secrecy arms race worth hundreds of billions. According to the Stanford AI Index 2025, the share of frontier AI systems released without full technical documentation more than doubled between 2022 and 2024. These systems are worth a fortune. The builders know exactly what they’re sitting on. They just won’t say.
The Race Nobody Is Watching
For most of AI’s recent history, research moved through public channels. Papers on arXiv. Conference presentations. Open benchmarks. That model is collapsing fast, and most people haven’t noticed.
World models are AI systems that don’t just predict the next word. They build internal maps of how things work. How a robot arm moves a box. How a car handles a crowded intersection. How a market reacts when rates change. These systems can simulate outcomes before taking action. That’s a fundamentally different kind of intelligence than what powers most chatbots.
According to MIT Technology Review, fewer than 30% of major AI systems released by frontier labs in 2024 included complete technical documentation. For world models specifically, the number is far lower. Companies like Physical Intelligence, Wayve, and Figure AI have collectively raised over $2 billion in the past two years with almost no published benchmarks or evaluation data, according to Crunchbase funding records and public reporting from TechCrunch.
This isn’t an accident. It’s a strategy.
Secrecy Is a Weapon, Not a Flaw
Most people look at this and see arrogance or fear. I see something else. I see companies that finally figured out what they’re worth.
For years, the AI industry ran on open publication. Share your research, build credibility, attract talent, win contracts. That worked when the gap between labs was small. Now that gap is not small. The top world model teams are working with capabilities that, if disclosed, would give competitors a one to two year head start. Why would they publish that?
Think about it like real estate. You don’t hand your competitor your deal flow. You don’t show them your underwriting model. You close deals and let your track record do the talking. The world model labs have adopted the same posture. Results are the pitch. The methodology stays locked.
According to a 2025 analysis by the Center for AI Safety, less than 12% of world model research from major commercial labs appeared in peer reviewed venues last year. That’s down from roughly 40% in 2021. The research isn’t disappearing. It’s moving behind closed doors, into patent filings, private demos, and government contracts.
This creates a massive information gap. And information gaps are where wealth transfers happen.
Here’s the brutal truth. If you’re an investor or builder trying to understand where AI is actually going, the public narrative is running two to three years behind the real research. The papers being published today reflect where the top labs were in 2023 and 2024. What they’re running right now won’t surface publicly until competitive pressure forces it out, or until someone gets acquired and the acquirer decides to brag.
If you’re running a media operation or content business in this space, you need to think about covering what you can’t see. Earnings calls. Patent filings. Regulatory submissions. That’s where the real signals leak. Tools like InVideo AI let you turn those weak signals into fast, clear video breakdowns before the mainstream press catches on.
What This Means for You
If you’re building a company, investing capital, or just trying to understand where money is flowing, the secrecy around world model research changes your playbook.
First, stop waiting for papers. Public research isn’t a leading indicator anymore. It’s a lagging one. The companies winning in world models aren’t the ones publishing the most. They’re the ones shipping products that use world models: robots, autonomous systems, financial simulation tools, and domain agents that can plan across time. Watch what gets funded. Watch what gets acquired. Watch what shows up in job postings at the big labs.
Second, focus on adjacent markets. You can’t build a competing world model in your garage. But you can build products on top of the APIs that do get released. You can serve the industries that world models will reshape, manufacturing, logistics, financial services, healthcare, before the reshaping happens. Early position in those markets is worth far more than late position.
Third, invest in your own knowledge stack. The information gap here is real. People who understand what world models actually do, not the hype version but the technical reality, will make better decisions for the next five years than everyone else. AppSumo is worth checking if you’re a builder who wants to stay ahead with the right tools without paying enterprise prices. But no deal replaces actually learning the material.
Here’s what I would do if I were starting fresh right now. I’d pick one industry that will be reshaped by physical AI, meaning world models plus robotics, and I’d become the most informed person in that intersection. Not a generalist. A specialist who can see the money before it moves.
The Bottom Line
World model companies aren’t keeping secrets because they’re afraid. They’re keeping secrets because the secrets are worth billions. The gap between what the top labs know and what the public knows is the largest it’s ever been in AI history. According to the Stanford AI Index 2025, overall AI R&D investment hit record levels in 2024 and 2025 while public disclosure rates dropped. That gap is an opportunity if you know where to look. Most people are reading yesterday’s news and calling it research. I’d rather follow the money than follow the papers.
Frequently Asked Questions
What is a world model in AI?
A world model is an AI system trained to understand how the physical or simulated world works, not just predict text or images. These systems can simulate outcomes before taking action, which makes them central to robotics, autonomous vehicles, and complex planning tasks. They’re fundamentally different from the large language models powering most chatbots today.
Why are world model companies keeping their research secret?
Because the research is worth too much to share. According to the Center for AI Safety, the share of commercial AI research appearing in public venues dropped sharply since 2021. Companies that publish their world model architecture give competitors a roadmap. The smarter play is to publish results and keep the methodology proprietary.
Which companies are building world models right now?
Physical Intelligence, Google DeepMind, Meta, Figure AI, Wayve, and OpenAI are among the most active in 2026. Most have raised hundreds of millions or more with limited public technical disclosure. You’ll get more signal from their funding announcements and product launches than from their research papers.
How does world model secrecy affect investors and builders?
It creates a large information gap between insiders and the public. Public narratives about AI capabilities are running two to three years behind what the top labs are actually doing. Smart builders and investors are watching job postings, patent filings, and acquisition activity instead of waiting for papers that may never come.
Is world model AI research actually slowing down?
No, it’s accelerating. But the acceleration is mostly happening in private. According to the Stanford AI Index 2025, overall AI investment hit record levels in 2024 and 2025. The research isn’t disappearing. It’s being treated as a competitive asset instead of a public good, and that changes everything about how you track it.
