Palo Alto Networks and NVIDIA Strengthen AI Governance Controls

Post Views: 6

Collaboration between Palo Alto Networks and NVIDIA aims to strengthen AI agent oversight through integrated security frameworks.

Palo Alto Networks and NVIDIA Collaborate on AI Agent Security

Palo Alto Networks and NVIDIA are collaborating to enhance oversight of AI agent operations through integrated security frameworks. The initiative focuses on managing agent behavior, network interactions, and identity governance using NVIDIA’s Open Agent Safety Platform reference design.

AI Agents and Security Challenges

AI agents, capable of executing tasks such as code generation, data retrieval, and tool utilization with minimal human intervention, require stringent security measures. Anand Oswal, EVP of AI and network security at Palo Alto Networks, emphasized that “security must control what an agent can do” in the company’s official statement.

“Security must control what an agent can do,” said Anand Oswal, EVP of AI and network security at Palo Alto Networks.

Deployment of Prisma AIRS AI Gateway

The collaboration involves deploying Palo Alto Networks’ Prisma AIRS AI Gateway on systems powered by NVIDIA Vera CPUs. This gateway enables centralized management of AI agent connections to models and tools, allowing organizations to monitor interactions, restrict access to approved resources, sanitize sensitive data, and track usage metrics.

OpenShell and Architectural Roadmap

The Vera-based deployment is part of a broader architectural roadmap. NVIDIA’s OpenShell software, already widely available, operates by isolating agents in controlled environments and enforcing policies that govern their access to files, networks, and system resources.

Runtime Security and Network Policies

Palo Alto Networks’ Prisma AIRS AI Runtime Security is deployed on NVIDIA BlueField data processing units, which are generally available. This setup separates security controls from the host system, enabling independent enforcement of network policies regardless of the application hosting the agent.

Integration and Future Plans

Integration with NVIDIA DOCA Argus allows monitoring of agent execution paths and tool interactions, ensuring compliance with access rules for networks and external APIs. Telemetry data is forwarded to Cortex XSIAM for analysis and response. A future integration with Palo Alto Networks’ Idira platform aims to introduce identity and secrets management for agent sandboxes, reducing reliance on persistent credentials tied to human users.

Conclusion

The partnership underscores efforts to align AI agent capabilities with organizational security requirements, ensuring access is strictly limited to task-specific needs.