Safeworld, a new startup emerging from stealth, is addressing the challenge of robot safety in the era of generative AI. Founded by Dr. Ding Zhao, a Carnegie Mellon AI safety expert, along with Kyle Wong and Simo Rachidi, Safeworld aims to ensure that humanoid robots powered by probabilistic generative AI models do not pose risks when deployed around humans. The difficulty lies not only in evaluating advanced AI behavior but also in building trust that these robots will operate safely in unpredictable, real-world environments.

The company has raised over $12 million in a seed round led by Shine Capital and a16z Speedrun, with additional investments from several venture firms and Carnegie Mellon. Safeworld is developing a simulation platform that models realistic human behavior and environments to test robotic control systems. This simulation approach resembles the validation processes used by autonomous vehicle firms but is considered more complex since robots must safely navigate unstructured settings with diverse safety standards, such as factory blind spots and human unpredictability.

Safeworld’s simulations replicate detailed scenarios, including human movements like tripping or carrying boxes, to test whether robots can recognize and appropriately respond to potential hazards. The founders believe roboticists will benefit from third-party validation to handle edge cases that are difficult to predict. Partnerships, such as with Gritt Robotics—which deploys robots alongside human workers on solar farms—highlight the need for empirical safety checks rather than purely theoretical guarantees, acknowledging the variability in human appearance and behavior that robots must accommodate.

Though still early, Safeworld is exploring whether to offer its product as a platform or a service, with confidence that their work will become essential as robots scale in use. Dr. Zhao predicts their business will be profitable because safety verification will be a prerequisite for all robot deployments, emphasizing that now is the critical time to establish safety standards before incidents occur in homes or workplaces. This initiative reflects the broader priority of embedding trust and accountability in generative AI-driven robotics from the outset.