A new startup called Safeworld has emerged from stealth with more than $12 million in seed funding to tackle one of the biggest challenges in robotics: ensuring that generative AI-powered robots won't harm people when they're deployed at scale. The company, founded by Carnegie Mellon University's Dr. Ding Zhao alongside startup veteran Kyle Wong and machine learning engineer Simo Rachidi, announced the funding round today, led by Shine Capital and a16z Speedrun, with backing from Box Group, Carnegie Mellon University Endowment, Innovation Endeavors, and SV Angel. The central problem Safeworld aims to solve is that generative AI models lack the predictability of traditional algorithms, making it difficult to guarantee a humanoid robot's safety in real-world settings.
Safeworld's approach involves testing robotic control systems inside simulations filled with realistic human models, running thousands of scenarios where simulated people interact with robots driven by their actual software. The company builds digital replicas of specific environments—like a factory's blind corner—using platforms such as Genesis or MuJoCo, then evaluates whether robots can detect humans carrying boxes, calculate proper stopping distances, or respond safely when people trip and fall. One early partner, Gritt Robotics, is working with Safeworld to develop safety simulations for robots that assist workers installing photovoltaic panels at industrial solar farms, where robotic arms must avoid hitting humans who might be kneeling, standing, crouching, running, or falling in various clothing and body configurations.
According to Dr. Zhao, the safety challenge combines two elements: "advanced generative AI probabilistic evals" to assess risk in probabilistic systems, and building trust sufficient for deployment. A16z Speedrun partner Jonathan Lai emphasized the urgency, stating that "the time to build an industry safety standard is now while robots are being designed and deployed," warning that waiting until robots are in households "colliding with kids and causing safety incidents" would be far too late. Gritt Robotics CTO Vishal Dugar explained that formally proving safety through mathematical equations isn't feasible for these systems, noting "it necessarily has to be done empirically."
Zhao argues that robots face greater safety challenges than autonomous vehicles because they operate in unstructured environments where each facility maintains different safety standards, and because human behavior is inherently unpredictable in ways that make testing difficult without simulation. The founders believe robot manufacturers will want third-party validation of their safety work, particularly to share information about safety cases among competitors. Zhao contends that many people are "underestimating how hard some of these edge cases are going to be to solve," distinguishing between controlled demonstration environments and real-world deployment scenarios where operators may have no prior robot experience. The company is still determining whether to offer its technology as a platform for external users or through a services model, though Zhao expressed confidence in the business case, predicting Safeworld will "probably be the first profitable company in this field" because "if anyone wants to deploy, they need to pay us to handle the situation."
The timing of this investment signals that venture capital now views safety infrastructure as essential rather than optional for the robotics industry. Whether centralized testing services can scale faster than internal safety teams at well-funded robot makers will determine if independent validators become gatekeepers or footnotes.