Agentic AI

NVIDIA and the Linux Foundation’s Open Secure AI Alliance is pushing for shared standards and tools to secure AI agents, with a focus on identity, permissions, isolation, logging, and incident sharing. For manufacturers, the key lesson is to ensure every deployed agent is fully auditable and independently inspectable, so a security incident can be investigated without relying entirely on the vendor.

As manufacturers deploy AI agents into production, they also expand their cyberattack surface, with prompt injection emerging as a major risk capable of manipulating agents into real-world operational actions. The key to safe adoption is treating AI agents as privileged users—limiting their access, requiring human approval for critical actions, and implementing strong monitoring, logging, and security controls from the outset.