Startup brings AI red-teaming to robots and self-driving cars

Laplace Intelligence is betting that artificial intelligence is the only practical way to safety-test other AI systems that operate in the physical world.

The company, founded about a year ago by CEO Howie Sun and chief scientist Henry Liu out of the University of Michigan, is building automated red-teaming for robotics and physical AI. Red-teaming is a security practice in which a system is deliberately stress-tested by attempting to make it fail.

Sun told TechCrunch the approach is motivated by lessons from incidents involving unpredictable AI agents, including a breach of pre-release models that OpenAI said was carried out through Hugging Face in July. "Imagine if something like that happened in the physical world," Sun said. "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's goal, Sun said, is to "help companies discover and fix problems before deployment — and, over time, build a shared foundation of evidence that customers, insurers, and regulators can trust."

The startup is among 20 companies competing in Startup Battlefield 200 at TechCrunch Disrupt in San Francisco this week, October 13-15, and was scheduled to present on the Disrupt Stage on Tuesday afternoon, October 13.

Analysis

Why This Matters

Physical AI systems such as robots and self-driving cars can cause real-world harm if they fail, making safety testing a matter of public safety, not just software quality.
Insurers, customers and regulators are likely to demand evidence of safety before physical AI is widely deployed, and a market for that evidence is only beginning to form.
The approach of using AI to test AI marks a shift in how safety assurance for autonomous systems may be produced at scale.

Background

Red-teaming has long been used in cybersecurity and military exercises, where an independent team attempts to break or defeat a system to expose weaknesses. As AI systems have become more capable, red-teaming has been adapted to probe large language models for harmful outputs or unexpected behavior. Laplace Intelligence is applying that same logic to physical systems, where a failure is not a harmful text response but a collision, an unsafe action, or a robot behaving unpredictably in the real world. The company emerged from the University of Michigan, and its founders are presenting their work at TechCrunch Disrupt as part of the Startup Battlefield competition.

Key Perspectives

Autonomous vehicle and robotics developers: They may gain a way to identify failures before deployment, reducing liability risk and strengthening safety cases to regulators and insurers. Insurers and regulators: They stand to benefit from standardized, shareable evidence about system safety, which could inform underwriting and certification decisions. Critics and skeptics: Automated red-teaming depends on the quality and imagination of the testing AI. A testing system that misses edge cases could create false confidence in the safety of physical AI systems.

What to Watch

Whether Laplace or similar startups can demonstrate a test failure that corresponds to a real-world hazard, not just a lab condition.
How regulators and insurers respond to evidence generated by AI red-teaming, and whether any adopt it as part of certification processes.
The outcome of the Startup Battlefield competition and whether Laplace secures funding or partnerships following its Disrupt presentation.

Analysis

Why This Matters

Background

Key Perspectives

What to Watch

Sources