Nutanix, Intel, and Cisco launch agentic AI security architecture for enterprises

This partnership establishes a standardized defense-in-depth model for autonomous systems that move beyond simple prompt-response interactions. By separating security into infrastructure, network, and control plane layers, the collaboration addresses the specific risk of agents misusing legitimate credentials during runtime loops or hallucinations. For the streaming and enterprise video sectors, this provides a blueprint for deploying autonomous agents that can safely interact with sensitive media databases and internal APIs without manual oversight. As organizations scale from experimental pilots to thousands of production agents, watch for whether this multi-vendor framework becomes the industry standard for managing non-human identity permissions and token budgets.

Nutanix has been building its AI infrastructure portfolio aggressively over the past year, positioning itself as a full-stack platform for enterprise AI workloads. In May 2025, Nutanix launched its AI Platform at the .NEXT conference, integrating GPU-accelerated compute with data services and model management to give enterprises a unified environment for training and inference. That platform expansion sets the stage for the Agent Gateway, which extends Nutanix's control plane into the emerging category of autonomous agent orchestration. Cisco, for its part, has invested heavily in securing AI workloads at the network layer. Cisco announced its Secure AI Factory reference architecture in June 2025, combining validated designs with NVIDIA GPUs and Cisco Silicon One networking to address data exfiltration and model poisoning risks in on-premises AI deployments. Intel has contributed hardware-rooted trust through its Xeon 6 processors, which include Intel TDX (Trust Domain Extensions) for confidential computing. Intel confirmed in March 2025 that Xeon 6 with TDX had been validated by major cloud providers for confidential AI inference workloads, establishing a hardware foundation that complements software-level agent permission controls.

The competitive landscape for agentic AI security is intensifying as enterprises move from pilot to production. Palo Alto Networks acquired Protect AI in April 2025 for approximately $700 million, signaling that AI model security is becoming a consolidation target among cybersecurity vendors seeking to cover the full AI lifecycle. Meanwhile, Microsoft announced in May 2025 that its Defender for Cloud service would extend agent identity governance to third-party AI agents running on Azure, directly competing with multi-vendor frameworks like the Nutanix-Intel-Cisco model for enterprise agent governance budgets. The Open Web Application Security Project (OWASP) also released its Top 10 for Large Language Model Applications in late 2024, which included excessive agency as a critical risk category, providing a shared taxonomy that vendors like Nutanix now reference when designing permission boundaries for autonomous agents.

Technical validation of multi-layer agent security remains limited, but early benchmarks suggest meaningful risk reduction. NVIDIA published a reference architecture in July 2025 for securing AI agent pipelines using network microsegmentation combined with hardware attestation on H100 and B200 GPUs, demonstrating that isolating agent runtime environments can reduce lateral movement attack surfaces by up to 90% in simulated enterprise networks. The Nutanix Agent Gateway approach mirrors this pattern by enforcing token consumption limits and identity verification at the control plane, though independent third-party testing of the combined Nutanix-Intel-Cisco stack has not yet been published. For streaming infrastructure operators evaluating for content moderation, metadata tagging, or CDN orchestration, the absence of standardized benchmarks means early adopters will need to conduct their own red-team assessments before production deployment at scale.