AI Testing AI: University of Michigan Spinout Laplace Intelligence Red-Teams Robots and Self-Driving Cars in Simulation

When OpenAI's pre-release agents slipped their sandbox and touched Hugging Face's infrastructure last summer, most of the industry read it as a software safety story. Howie Sun read it as a preview of something worse. "Imagine if something like that happened in the physical world," the CEO and co-founder of Laplace Intelligence told TechCrunch. "AI agents are moving from computers into the real world, and we don't really want physical AI to fail in the real world."

Laplace, based in Ann Arbor, Michigan, was founded a year ago by Sun and chief scientist Henry Liu — his then-PhD adviser — as a spinout from the University of Michigan. Sun spent eight years in academia researching how to challenge autonomous driving AI in simulation, and in 2023 the pair published a paper on the subject that made the cover of Nature. That is when the calls started coming in from autonomous vehicle companies, and Sun decided to turn the research into a product.

The company is named for "Laplace's demon," the 1814 thought experiment in which a being that knew the exact state of everything in the universe could predict the future with total precision. "We are essentially using simulation to predict the real world's future," Sun said. At the core is a high-fidelity simulation environment reconstructed from the real world, built on a post-trained version of Nvidia's Cosmos world-simulation platform. It blends real-world map data — a Street View image can seed an entire street, generated true to scale — and developers can alter scenes with language prompts, adding afternoon traffic drawn from live TomTom data.

What makes the environment unusual is that everyone in it reacts. In a demo, Sun directed the test agent vehicle to swerve into oncoming traffic — and the oncoming car moved out of the way to avoid the collision, behavior static generative world models typically lack. "It's truly an interactive simulation environment," Sun said. "Everyone is reacting to everything… This is more like a world model with interactive participants."

The real product, though, is the agents on top of the simulator. Laplace is building AI agents that automatically design corner cases — the rare, tricky scenarios that trip up AI systems — drawing on sources like news reports and police collision records, moving humans out of the loop. Today, AV developers mostly replay recorded footage or rely on human imagination to invent edge cases. For generalized robots, Sun argues, that will not scale: "There's less data on generalized robotics. No one knows what the edge cases will be in the real world."

The company already has revenue and says it is modestly profitable, driven mostly by autonomous driving work. Customers include Japanese trucking manufacturer Isuzu, Texas-based autonomous trucking firm Bot Auto and Nvidia; together with Nvidia and T-Mobile, Laplace is working on a free smart-city pilot in San Jose, California, and is in early stages with partners building humanoid robots for warehouses. Most contracts are project-based today, with the company working to convert them into recurring revenue.

The shadow competitor is Applied Intuition, the late-stage simulation startup that has raised $1.2 billion at a $15 billion post-money valuation — and, like Laplace, counts Isuzu as a customer. Sun is not discouraged. He believes AI testing AI is "a new paradigm that changes the testing of physical AI entirely," turning verification into a dynamic, interactive process — and that the methodology itself is the moat. Laplace is one of 20 startups presenting at TechCrunch Disrupt's Startup Battlefield 200 this week.

If physical AI is going to deploy at scale, someone will have to certify it — and customers, insurers and regulators will want evidence they can trust. Simulation-based red-teaming is a credible candidate for that compliance layer. For now, the performance claims are Laplace's own; the near-term test is whether "AI testing AI" can win contracts against a rival with a $15 billion valuation and an eight-year head start on enterprise sales.