Safeworld raises $12M to safety-test gen AI robots
Original: Can Safeworld convince people that gen AI robots won’t hurt them?
Why This Matters
As humanoid robots enter factories and homes, safety certification gaps could stall the entire sector.
Safeworld, founded by CMU's Dr. Ding Zhao, Kyle Wong, and Simo Rachidi, exits stealth with $12M+ seed funding led by Shine Capital and a16z Speedrun to build safety evaluation tools for generative AI-powered robots.
The shift toward generative AI as the brain of humanoid robots creates a thorny problem: unlike traditional algorithms, gen AI systems are probabilistic and unpredictable. Safeworld wants to be the company that stress-tests them before they hurt anyone.
Founded by Dr. Ding Zhao — director of Carnegie Mellon's Safe AI lab — alongside startup veteran Kyle Wong and ML engineer Simo Rachidi, Safeworld's approach centers on simulation. The company builds digital replicas of real-world environments — a factory blind corner, a warehouse aisle — using physics engines like Genesis or MuJoCo, then runs thousands of scenarios with realistic human models against the robot's actual software. Edge cases like a person tripping, or carrying boxes that obscure their silhouette, are tested at scale without putting anyone at risk.
The $12M+ seed round includes Box Group, CMU Endowment, Innovation Endeavors, and SV Angel alongside the lead investors. a16z Speedrun partner Jonathan Lai framed the urgency bluntly: "By the time you have robots in households colliding with kids and causing safety incidents, that's way too late."
Zhao identifies two distinct hurdles: evaluating probabilistic AI risk (how do you underwrite a system that doesn't always behave the same way?), and building enough trust for operators to actually deploy robots. Each facility adds complexity, since safety standards vary by site.