Securing AI Agents at Scale with NVIDIA

AI agents are evolving into autonomous systems capable of reasoning, planning, and acting with minimal human oversight, executing tasks such as writing code and accessing enterprise data. Traditional prompt guardrails are insufficient for these agents, as their ability to adapt to blocked paths requires security controls external to the agent itself. NVIDIA provides the computing platform and runtime foundation, while Palo Alto Networks contributes AI security capabilities to govern interactions, protect workloads, and manage identity and access.

The collaboration introduces three security areas: AI governance via Prisma AIRS AI Gateway, runtime protection through Prisma AIRS AI Runtime Security, and Agent Identity Security powered by IDIRA. The AI Gateway enforces real-time governance by verifying agent identities, restricting access, redacting sensitive data, and inspecting interactions to block unauthorized actions before they reach critical systems.

Runtime security is extended to AI workloads using the Palo Alto Networks AI Runtime firewall, which operates natively on NVIDIA BlueField DPUs to protect front-end network traffic in AI factories and at the edge. This firewall creates a tamper-proof boundary against application-layer attacks, leveraging the NVIDIA Sentry reference design to deliver accelerated runtime security for agentic systems.

Identity management becomes critical as agents operate autonomously, requiring controls independent of human credentials. The partners plan to integrate NVIDIA OpenShell with upcoming Agent Identity Security capabilities, including identity brokering on NVIDIA BlueField, to enforce least-privilege access and prevent privilege escalation across enterprise systems.