The landscape of artificial intelligence is undergoing a profound transformation. No longer confined to passive assistants, AI agents are rapidly evolving into sophisticated, autonomous systems capable of independent reasoning, planning, and execution. This seismic shift necessitates a fundamental re-evaluation of security models, moving beyond traditional guardrails to robust, external control mechanisms. In response, Palo Alto Networks, a global leader in cybersecurity, and NVIDIA, the titan of accelerated computing, have announced a significant expansion of their collaboration, aiming to establish a secure foundation for the burgeoning field of agentic AI.
This strategic partnership is designed to address the critical security challenges posed by AI agents that can write and execute code, access sensitive enterprise data, interact with external tools and APIs, and persistently pursue goals with minimal human oversight. The inherent nature of these autonomous agents, their ability to dynamically adapt and find alternative pathways to achieve objectives, renders conventional, built-in security measures insufficient. The core principle driving this collaboration is the understanding that while AI "harnesses" can guide an agent’s actions, comprehensive security must dictate what an agent can do, requiring an authority that operates independently of the agent itself.
The Imperative for Agentic AI Security: A Paradigm Shift
The ascent of agentic AI marks a pivotal moment, signaling a departure from reactive, command-driven interactions to proactive, goal-oriented operations. These agents are engineered to be problem-solvers, capable of complex reasoning and strategic planning. When faced with an obstacle, an autonomous agent doesn’t simply halt; it analyzes the situation, formulates new strategies, and iteratively works towards its defined objective. This self-directed problem-solving capability, while the source of their immense power, also presents significant security vulnerabilities.
Traditional security models, often built around perimeter defenses and static rule sets, struggle to keep pace with the dynamic and fluid nature of autonomous agents. The agents’ ability to learn, adapt, and interact with a vast array of digital resources means that threats can emerge from unexpected vectors and evolve at an unprecedented rate. Relying solely on the internal "guardrails" of an AI model, which are essentially instructions embedded within its programming, is akin to providing a driver with a speed limit sign but no brakes. The agent might understand the instruction, but its inherent drive to achieve its goal could override the intent if not properly constrained.
The critical distinction lies in the concept of "guidance" versus "control." While prompts and internal parameters can guide an agent’s behavior, they do not provide the robust, external enforcement necessary to prevent malicious or unintended actions. True security for agentic AI demands an authoritative layer that operates outside the agent’s direct purview, capable of preemptively blocking unauthorized access, scrutinizing every interaction, and enforcing granular access policies. This external authority acts as the ultimate arbiter, ensuring that agents operate within defined ethical and security boundaries.
Palo Alto Networks and NVIDIA: A United Front for Secure Agentic AI
The expanded collaboration between Palo Alto Networks and NVIDIA is built upon a shared vision of enabling the widespread adoption of agentic AI without compromising enterprise security. NVIDIA, with its unparalleled expertise in accelerated computing, secure runtime foundations, and reference system designs, provides the essential infrastructure for agentic AI to flourish. This includes powerful GPUs, specialized processors like the NVIDIA Vera CPU, and advanced networking solutions such as the NVIDIA BlueField DPU.
Palo Alto Networks, in turn, brings its deep bench of AI security capabilities, honed through years of defending against sophisticated cyber threats. Their expertise lies in governing AI interactions, protecting AI workloads from compromise, and rigorously controlling the identity and access of AI agents. Together, they are developing a comprehensive security framework that addresses the multifaceted challenges of agentic AI.
This ambitious initiative focuses on three key areas, forming the pillars of their integrated security approach:
- AI Governance with Prisma AIRS AI Gateway: This component is designed to establish real-time policy enforcement and control over AI agent activities, ensuring that they operate within predefined governance parameters.
- Runtime Protection with Prisma AIRS AI Runtime Security: This layer provides continuous monitoring and defense against threats at the runtime of AI agents, safeguarding against malicious actions and unauthorized access during operation.
- Agent Identity Security powered by IDIRA: Recognizing that identity is paramount for controlling autonomous entities, this aspect focuses on establishing secure identities for AI agents, enabling granular access control and accountability.
These three pillars are being integrated into the NVIDIA Open Agent Security Platform reference design, a testament to the depth and breadth of this partnership.
Prisma AIRS AI Gateway on NVIDIA Vera CPU: The First Line of Defense
At the forefront of this security architecture is the Prisma AIRS AI Gateway, which will operate on NVIDIA Vera CPU systems. This strategic placement brings AI governance and enforcement capabilities directly to the infrastructure designed to host agentic workloads. The NVIDIA Vera CPU, a powerful Arm-based processor designed for data center efficiency and security, provides a robust and performant platform for the AI Gateway.
The AI Gateway acts as a critical real-time enforcement point for all aspects of AI security: governance, runtime, and identity. Its core functions are designed to proactively prevent security breaches before they can impact critical systems. This includes:
- Continuous Identity Verification: The Gateway rigorously verifies the identity of every AI agent attempting to access resources, ensuring that only authorized entities can proceed.
- Strict Least-Privilege Access Enforcement: It enforces the principle of least privilege, granting agents only the minimum permissions necessary to perform their designated tasks. This significantly reduces the attack surface and limits the potential damage from compromised agents.
- Sensitive Data Redaction: Before sensitive data can be exposed to an agent, the Gateway inspects and redacts it, preventing inadvertent disclosure of confidential information.
- Deep Interaction Inspection: Every interaction initiated by an AI agent is deeply inspected. This allows for the identification and blocking of suspicious or unauthorized actions in real-time, preventing them from reaching critical host systems or cloud endpoints.
By embedding these enforcement capabilities directly into the infrastructure, the Prisma AIRS AI Gateway creates a formidable barrier, ensuring that agentic AI operates within a secure and controlled environment from its inception. This approach is crucial for enterprises that are increasingly deploying AI agents to manage complex workflows and access sensitive business data.
Announcing Prisma AIRS AI Runtime Security on NVIDIA BlueField: Securing the AI Factory
Extending Zero Trust protection to the very fabric of the AI operational environment, Palo Alto Networks is announcing Prisma AIRS AI Runtime Security, which will run natively on NVIDIA BlueField DPUs (Data Processing Units). The NVIDIA BlueField DPU is a specialized processing unit designed to offload, accelerate, and isolate networking, storage, and security tasks from the host CPU, significantly enhancing the performance and security of data center infrastructure.
The AI Runtime firewall, accelerated and secured by the BlueField DPU, forms a tamper-proof boundary around the AI factory and the AI Edge. This boundary is designed to be immune to application-layer attacks, a common threat vector in modern IT environments. By leveraging the NVIDIA Sentry reference design, a framework for building secure and performant data center infrastructure, Prisma AIRS AI Runtime Security delivers a robust defense posture.
Key benefits of this integration include:
- Enhanced Network Security for AI Workloads: The BlueField DPU’s offload capabilities allow the AI Runtime firewall to inspect and secure network traffic at line rate, without impacting the performance of AI training and inference.
- Zero Trust Enforcement at the Network Edge: By deploying security functions directly onto the DPU, Zero Trust principles are applied at the network edge, ensuring that every connection and every packet is authenticated and authorized.
- Tamper-Proof Security Boundary: The inherent isolation and security features of the BlueField DPU create a highly resilient security boundary that is extremely difficult for attackers to compromise.
- Protection Against Application-Layer Attacks: The advanced inspection capabilities of the AI Runtime firewall, coupled with the DPU’s performance, enable effective defense against sophisticated application-layer threats that target AI systems.
This deployment on the BlueField DPU signifies a proactive approach to securing the entire lifecycle of agentic AI, from its development and training within the AI factory to its deployment and operation at the AI Edge.
IDIRA Agent Identity Security: The Foundation of Trust for Autonomous Agents
As AI agents move towards greater autonomy, the concept of identity becomes paramount. Traditional identity and access management systems are designed for human users, and their direct application to autonomous agents can be complex and insecure. This is where IDIRA Agent Identity Security plays a crucial role.
The partnership recognizes that for agents to operate safely and responsibly, their identities must be securely managed and their access strictly controlled. Looking ahead, the collaboration plans to integrate NVIDIA OpenShell, an open-source framework for building and deploying AI applications, with upcoming Agent Identity Security capabilities from Idira.
This integration will focus on:
- Identity and Secrets Management for OpenShell Sandbox: Securely managing the identities and secrets (e.g., API keys, credentials) used by agents within the OpenShell sandbox environment. This ensures that agents have access only to the secrets they require for their operations.
- Agent Identity Broker Service on NVIDIA BlueField: Establishing an identity broker service that runs on NVIDIA BlueField DPUs. This service will act as a central authority for verifying agent identities and brokering access to resources.
By leveraging these capabilities, organizations will be able to apply robust, identity-based controls and least-privilege access to AI agents without the need for persistent, human-managed credentials. This is a significant step towards enabling secure, scalable deployment of autonomous AI.
The Path Ahead: Comprehensive Security for Enterprise Agents at Scale
The ongoing evolution of AI agents demands a security strategy that can scale with their increasing autonomy and complexity. The combined efforts of Palo Alto Networks and NVIDIA are meticulously designed to address this challenge, ensuring that enterprises can harness the power of agentic AI with confidence.
The Prisma AIRS AI Gateway will evolve to serve as a critical enforcement point for Agent Identity Security powered by IDIRA. This means that the governance and policy enforcement capabilities of the Gateway will extend to the management of agent identities, ensuring that access decisions are based on verified identities and predefined security policies.
Furthermore, the expansion to include Agent Identity Broker support directly on NVIDIA BlueField DPUs is a game-changer. This capability will provide independent identity verification and access governance for agents, decoupled from the host environment. This separation of concerns enhances security by reducing the attack surface and preventing potential compromise of the host from affecting agent access controls.
Collectively, these advancements are engineered to achieve several critical security objectives:
- Prevent Unauthorized Execution: By rigorously verifying agent identities and enforcing strict access controls, the risk of unauthorized code execution is significantly mitigated.
- Reduce Standing Access: The principle of least privilege, coupled with dynamic access management, ensures that agents only have access to the resources they need, when they need them, thereby reducing the potential for misuse or compromise of broad access privileges.
- Stop Privilege Escalation: By continuously monitoring agent activity and enforcing granular permissions, the security framework aims to prevent agents from escalating their privileges beyond their authorized scope.
As AI agents become more sophisticated and integrated into enterprise workflows, the need for consistent and reliable control over their connectivity and access to resources becomes paramount. The collaboration between Palo Alto Networks and NVIDIA is strategically placing these essential controls closer to where agentic workloads are executed. Palo Alto Networks provides the overarching security policy, governance frameworks, and enforcement mechanisms that enterprises require to manage these powerful new systems effectively.
The integration of Palo Alto Networks’ security solutions with NVIDIA’s advanced hardware and reference architectures represents a significant leap forward in securing the enterprise AI factory and AI edge. This ensures that the benefits of AI can be realized without introducing unacceptable security risks.
For organizations looking to embrace the future of AI, the question is not just about what agents can do, but whether they can be trusted to operate safely and securely at scale. The ongoing efforts of Palo Alto Networks and NVIDIA are laying the groundwork for this trusted future, where autonomous AI can be deployed with the same level of confidence as any other critical enterprise system.
Palo Alto Networks has been recognized for its pioneering work in AI application security, being named a Market Shaper in the recent Gartner® Emerging Market Quadrant for AI Application Security – Established Vendors, September 2026. This validation underscores the company’s commitment to staying ahead of the curve in this rapidly evolving domain. The continued collaboration with NVIDIA signifies a unified commitment to shaping the future of secure, autonomous AI, ensuring that innovation and security advance hand-in-hand.
