AI Agents Can Shape the Physical World. We Can Secure Their Actions.

AI agents can be secured. In Physical AI, we must also secure what happens when their decisions become a machine’s actions.
 
For years, the boundary around many robots was easy to see: a fence. Machines here. People there.
 
Physical AI is changing that picture.
 
Robots are entering spaces shared with people. Autonomous vehicles must respond to roads that change from one moment to the next. NVIDIA is advancing AI agent safety at a time when the industry is building autonomous, safety-critical machines for complex environments.
 
As AI agents become part of these systems, their decisions can influence a vehicle or robot’s next action. A malicious instruction hidden in data, a compromised tool, or overly broad permissions can redirect an otherwise legitimate operation. The agent may stay within its assigned permissions while the machine moves beyond its intended safety boundary.
 
We can build for that reality.
We can verify identity.
We can limit authority.
We can test what an agent does when its inputs are manipulated.
We can detect when those actions change machine behavior.
And we can intervene when a safety boundary is at risk.
 
This is a responsibility for cybersecurity, working alongside the engineers who design these machines for safety. At VicOne, we are advancing this work across automotive and robotics. In vehicles, we help secure onboard AI agents. In robots, we help secure AI models and AI agents, test how attacks could change robot behavior before deployment, and detect related threats during operation.
 
We applaud NVIDIA’s launch of its Open Agent Safety Platform and support its effort to make agent actions secure and governable. For vehicles and robots, those safeguards must also account for what the machine does in the physical world. That is the contribution VicOne is committed to making.