SafeWorld Emerges From Stealth With $12.2M to Simulate Robot Safety Scenarios

SafeWorld this week emerged from stealth with $12.2 million in seed funding, betting that the next bottleneck in robotics isn't capability — it's proving that capable robots are safe to deploy around people.

The Oakland, Calif.-based startup aims to make safety testing on AI-powered robots easier with its simulation technology, offering what it calls a scalable way to test how robots behave around people in rare, dangerous, and unexpected situations — without putting anyone at risk.

Why now

Ding Zhao, director of the Safe AI Lab at Carnegie Mellon University, frames the timing around a structural shift in automation. Robots have been in use for maybe 100 years, but AI changed the operating conditions.

"We started the company today because, when AI came, it actually brought automation to the next level. The robots, for the first time, were allowed to go out of the cage and share the same space with humans."

That shift — from caged, fixed automation to shared-space machines — is what creates the testing problem. SafeWorld's platform can work with various robot embodiments, including robot arms, humanoids, and mobile robots, and the company already reports customers in industrial, manufacturing, logistics, construction, and some home use cases.

The funding round was co-led by Shine Capital and a16z Speedrun, with participation from Box Group, Carnegie Mellon University Endowment, Innovation Endeavors, SV Angel, and other venture capitalists and angel investors.

How the platform works

Teams can build scenarios directly in the browser — drawn from past incidents, safety standards, and robot logs — without simulation expertise. SafeWorld then runs the robot through thousands of variations with realistic, reactive human motion and measures safety performance.

"We are providing robot safety simulation software that robotics companies or enterprises that are building robots can use to better provide evidence around safety, to test different types of safety scenarios before they happen in the real world, or test things that are effectively too rare, too dangerous, or too expensive to test in the real world," Wong said.

That last category matters most. "These are safety scenarios that have not existed yet in the world," Wong said — meaning there's no wide dataset to train against.

Because every software update and new environment can introduce new risks, the platform supports continuous re-testing rather than a single pre-launch validation pass. Wong acknowledged that applying simulation specifically to safety — rather than general development — brings new requirements. "Where that breaks down and gets different is really in the nuance of the feature and functionality," he said, pointing to incomplete standards and varying risk assessments.

Timing the safety investment

SafeWorld currently works with about half a dozen companies, two-thirds of them publicly traded, according to Wong. His rule of thumb: if deployment is under one year out, start thinking about safety now; beyond a year, it's probably too early.

The calculus also differs by company type. A company like Caterpillar, which already owns sophisticated hardware working in the world and wants to add AI capabilities, needs safety on Day 1. Software-first companies face the problem later — but getting ahead of it can still streamline deployment.

Thomas Tang, CEO of Anyware Robotics, is one customer already on board. "Safety has been foundational to Anyware Robotics from Day 1," he said, adding that advanced simulation tools give his team a scalable way to test challenging scenarios and strengthen safety processes.

The trust argument

SafeWorld plans to use the funding to deepen enterprise adoption and promote safety research across the industry.

"We're really trying to go super deep in terms of enterprise adoption and going super deep with some of our enterprise customers," Wong said. "On the marketing side, we're trying to do a lot in terms of promoting different research, whether that's funding research or promoting research, and open-source work that is at the frontier of AI safety."

The co-founders also make a pointed case for why safety validation shouldn't be built in-house every time. "Part of the reason why I think a third party is very helpful here is that there is an engineering part around safety, and then there is a trust part of safety, and you need both of those to deploy," Zhao said.

"This is really a deployment problem. So, having a third-part company to work on this is really necessary to win trust not only for the company itself, but also for the whole industry," he added.