Figure Joined NVIDIA's New Safety Push for AI Agents
On Sept. 28 NVIDIA launched an Open Agent Safety Platform with more than 100 partners. Figure signed on. It has not said which of its robots will use the tools.
So basically: Figure signed onto a new NVIDIA safety effort for AI agents on Sept. 28. The tools are meant to limit what an agent can do. Figure has not said how its robots will use them.
The announcement
NVIDIA CEO Jensen Huang announced the Open Agent Safety Platform on Sept. 28. He said more than 100 companies are in it. Figure CEO Brett Adcock tied the news to robots around people. His line: humanoid robots will soon be in the home and workplace, and they have to be safe and trusted.
The same day, NVIDIA published OpenShell 0.1.0. It is open source software that runs an agent in a sandbox and sets which files, tools, and network calls that agent is allowed to use. The limits sit outside the model. A bad decision from the model does not automatically turn into an action.
What Figure did not say
Adcock’s post confirms Figure is part of the effort. It does not name a robot, a factory, or a date. Humanoids Daily looked at NVIDIA’s partner graphic and said Figure was the only company on it that builds only humanoids. Gecko Robotics and Skild AI are on the graphic too. Figure has not said what it will actually plug in.
My take
A name on a partner list is not a safety system inside a robot. The useful part is where the limits sit: outside the model, on the actions. That is the setup you want for a machine that can walk into a room. Figure still has to show the software running on one of its own robots.
So basically — pass it on.