Safeworld emerges from stealth to safety-test AI robots

Safeworld, a startup co-founded by Dr. Ding Zhao, who directs the Safe AI Lab at Carnegie Mellon University, startup executive Kyle Wong and machine learning engineer Simo Rachidi, has emerged from stealth with a seed round of more than $12 million, led by Shine Capital and a16z Speedrun, with additional investment from Box Group, Carnegie Mellon University Endowment, Innovation Endeavors and SV Angel.

The company is tackling a problem created by the shift to generative AI in robotics: a generative AI model driving a robot behaves unpredictably, unlike traditional algorithms, making it harder to guarantee a humanoid robot is safe. Zhao frames the challenge as two linked problems — probabilistic evaluations that can underwrite the risk of an inherently probabilistic system, and the trust needed to actually deploy a robot. a16z Speedrun partner Jonathan Lai said the industry needs a safety standard now, before robots start causing safety incidents in households.

Safeworld evaluates a robot’s control system inside simulations populated with realistic human models. Zhao argues this is harder for robots than for self-driving cars like those from Tesla and Wayve, because robots operate in unstructured environments where each facility sets its own safety rules. Wong gave the example of a blind factory corner: what stopping distance keeps a robot from colliding with a worker carrying boxes? Safeworld builds a digital replica of the corner in a simulator such as Genesis or MuJoCo, inserts a simulation of the robot driven by its real software, and runs thousands of scenarios of human models encountering it — including people tripping and falling, since repeatedly testing that scenario with a physical robot is itself difficult to arrange safely.

The founders believe robot makers will want third-party validation beyond their own internal tools, partly so they can share safety findings with competitors without exposing proprietary systems. “It is not the robot in the vacuum, in the demo, that we are worried about,” Zhao said. “It is the robot that is deployed at scale, with people who potentially never operated a robot before.” Vishal Dugar, CTO of Gritt Robotics — which builds AI systems for robots that help install photovoltaic panels at industrial-scale solar farms — is partnering with Safeworld. Dugar said safety can’t be proven through equations alone; it has to be verified empirically, given how many different postures and appearances humans present on a job site.

It’s still early for both Safeworld and generative AI in robotics, and the company hasn’t settled on whether to sell a platform or operate as a service. Zhao predicts Safeworld will likely be the first profitable company in the field, since any company that wants to deploy a robot will need to pay someone to handle its safety case.

Why it matters

As generative AI moves from chatbots into physical robots working alongside people, third-party safety validation could become a prerequisite for deployment — shifting how manufacturers prove their machines are safe from one-off engineering sign-offs to standardized, shareable evaluations.