Security enters the agents stack
NVIDIA has announced Open Agent Safety Platform, an open software platform and a reference design intended to strengthen the security, governance, and control of AI agents from testing through deployment.
Why it’s needed
The more an agent can use tools and take actions, the more important it becomes to limit its privileges and behavior. NVIDIA therefore aims to integrate controls at both the software and hardware levels.
A problem destined to grow
Agentic security could become a standard component of AI infrastructure. If agents become a permanent part of enterprises, it will not be enough to assess their intelligence: it will be necessary to be able to observe, govern, and contain what they do.
Source: NVIDIA Newsroom, 28 September 2026.