Laplace Intelligence Wants AI to Red Team Robots Before Launch

Laplace Intelligence, a University of Michigan spinout based in Ann Arbor, builds automated red teaming for physical AI. Its AI agents and interactive simulations deliberately try to make robots and self driving vehicles fail before they are deployed.

CEO Howie Sun told TechCrunch the bet is that the only way to keep physical AI safe at scale is to have AI test it. The company says it is already profitable and pitched at Startup Battlefield during TechCrunch Disrupt.

Why it matters: as robots run on large learned models, failure modes get harder to predict. Safety testing is becoming a product category of its own.