Accelerating agent intelligence: Cisco Secure AI Factory expands with NVIDIA and VAST

[ad_1] Synthetic intelligence is getting into a brand new part. The dialog is shifting from mannequin innovation to operational actuality. Organizations are discovering that constructing AI fashions is commonly the simplest a part of the journey. Working these fashions reliably, securely, and at scale throughout enterprise environments is the place complexity emerges.

Many AI initiatives decelerate, not as a result of groups lack GPUs, knowledge, or expertise, however as a result of there is no such thing as a unified working sample that safely connects all of these parts into manufacturing. AI methods will not be single functions. They're distributed ecosystems of knowledge pipelines, inference providers, orchestration layers, and more and more, autonomous brokers interacting with enterprise methods in actual time.

Cisco Safe AI Manufacturing facility with NVIDIA is constructed round a easy however transformative thought. AI have to be handled as an end-to-end system. Efficiency, knowledge readiness, cloud-native operations, and safety can't be designed individually. They have to be engineered collectively from the start.

At VAST Ahead 2026, we're demonstrating how that precept interprets right into a working safe AI knowledge platform. This isn't a future idea or hypothetical structure. It's a actual, deployable reference implementation constructed utilizing NVIDIA accelerated computing infrastructure and software program, VAST knowledge providers, Cisco infrastructure, the Isovalent Enterprise Platform based mostly on Cilium and Tetragon, and Cisco AI Protection. It displays a repeatable option to operationalize AI right now whereas persevering with to evolve towards deeper integration over time.

The brand new actuality of enterprise AI The rise of retrieval-augmented technology (RAG) and agent-driven functions is basically reshaping how organizations work together with their knowledge. AI methods are not remoted workloads. They repeatedly retrieve data, trade context between providers, and execute automated actions throughout enterprise environments.

This transformation introduces a brand new sort of operational problem. The assault floor expands dramatically as AI workloads generate fixed east-west site visitors inside Kubernetes clusters. Runtime habits turns into extra dynamic as containers load libraries, execute helper processes, and work together with exterior providers. On the identical time, fashions and brokers introduce dangers that conventional safety instruments have been by no means designed to handle, together with immediate injection, delicate knowledge leakage, and uncontrolled device execution.

Enterprise leaders will not be asking whether or not these dangers exist. They're asking whether or not AI might be trusted to ship measurable outcomes with out exposing the group to unacceptable operational or regulatory publicity. The reply lies in designing AI platforms the place safety is inseparable from efficiency and scalability.

Constructing the platform from the information outward Each efficient AI system begins with knowledge that's accessible, constant, and instantly usable. VAST Information Platform and VAST InsightEngine rework enterprise knowledge into an lively participant in AI workflows slightly than a passive storage layer. By automating ingestion, indexing, and retrieval pipelines, the platform permits enterprise knowledge to turn into a dependable context for AI methods with out the delicate and sophisticated knowledge engineering pipelines that always sluggish innovation.

Working this knowledge intelligence layer on Cisco UCS and NVIDIA accelerated computing, software program, and networking permits the platform to maneuver past experimental deployments. It creates a repeatable constructing block that organizations can deploy throughout environments with constant efficiency and lifecycle administration. Manufacturing AI requires this degree of operational self-discipline. With out it, scaling AI turns into unpredictable and troublesome to control.

The place safety should dwell in trendy AI platforms Probably the most vital shift in AI safety is the placement. Safety can not focus solely on defending the community perimeter or scanning container pictures earlier than deployment. In AI knowledge platforms, nearly all of danger now exists inside Kubernetes clusters and inside AI software interactions themselves.

The primary essential problem is controlling east-west site visitors. AI microservices talk repeatedly as retrieval pipelines, embedding providers, and inference engines trade knowledge. With out robust segmentation, unintended service reachability can emerge as clusters scale, permitting lateral motion throughout workloads.

The Isovalent Enterprise Platform based mostly on Cilium addresses this problem by imposing identity-based community insurance policies instantly inside Kubernetes. As an alternative of counting on fragile, IP-based guidelines, insurance policies comply with workload identification as providers scale, migrate, or restart. This ensures that solely licensed providers talk with each other whereas sustaining excessive efficiency by means of eBPF-accelerated networking. The result's constant enforcement of least-privileged communication throughout the cluster.

Nevertheless, community segmentation alone can not detect surprising habits inside containers. AI workloads incessantly execute processes, entry delicate recordsdata, and dynamically load instruments and libraries. Even when community communication is restricted, compromised workloads can nonetheless behave unpredictably at runtime.

Isovalent Enterprise Runtime Safety, constructed on Tetragon, addresses this second layer of danger. By offering kernel-level observability of course of execution and file exercise, it permits operators to grasp what workloads are doing inside containers. Suspicious habits might be recognized early, serving to organizations examine and reply earlier than points escalate.

Collectively, these capabilities create a significant and enforceable Kubernetes safety posture. They management how providers talk and supply visibility into how workloads behave throughout execution.

Extending safety to the AI layer itself The quickest rising danger floor in AI environments sits on the mannequin boundary. Fashions and brokers function in dynamic environments the place consumer prompts, enterprise knowledge, and exterior instruments intersect. Conventional safety instruments weren't constructed to detect manipulation of AI interactions or unsafe agent habits.

Cisco AI Protection brings safety instantly into the AI software layer. It helps organizations analyze mannequin elements for vulnerabilities, apply runtime guardrails to prompts and responses, and monitor how fashions work together with instruments and knowledge sources. This gives visibility into how AI methods behave and helps cut back the chance of enterprise knowledge or automated agent actions creating unintended publicity.

With this layer in place, safety spans the total lifecycle of AI workloads, from infrastructure and knowledge to Kubernetes operations and AI software habits.

Demonstrating the safe AI knowledge platform in motion At VAST Ahead 2026, we're exhibiting this structure working as an entire and purposeful resolution. Enterprise knowledge is reworked into AI-ready context by means of the VAST pipeline. The platform runs on Cisco infrastructure aligned to Cisco Safe AI Manufacturing facility with NVIDIA design rules. Kubernetes east-west site visitors is segmented utilizing the Isovalent Enterprise Platform based mostly on Cilium, whereas runtime habits is monitored utilizing Isovalent Enterprise Runtime Safety constructed on Tetragon. The AI interplay layer is protected utilizing Cisco AI Protection.

This isn't a theoretical blueprint. It's a dwell, deployable reference structure that prospects can implement right now whereas persevering with to evolve towards deeper integration and automation.

The shift towards safe AI outcomes Crucial lesson rising from enterprise AI adoption is that safety can't be measured by the variety of controls deployed. It have to be measured by the flexibility to function AI safely and confidently at scale.

A safe AI knowledge platform permits organizations to ship this consequence by guaranteeing:

AI pipelines stay remoted throughout groups and workloads East-west site visitors inside Kubernetes is managed and observable Runtime habits inside containers is monitored and understood Fashions and agent interactions are shielded from rising AI-specific threats

When these parts are designed collectively, organizations acquire the arrogance to scale AI initiatives throughout departments, functions, and enterprise items.

The way forward for accountable AI operations Cisco Safe AI Manufacturing facility with NVIDIA represents a blueprint for a way enterprise AI might be constructed shifting ahead. It brings efficiency, knowledge intelligence, cloud-native operations, and AI-native safety collectively in a unified operational sample.

Organizations not want to decide on between velocity and security. They'll deploy AI methods which are each revolutionary and reliable, permitting them to maneuver from experimental initiatives to manufacturing AI providers that ship actual enterprise affect.

If you're attending VAST Ahead 2026, we invite you to expertise this resolution firsthand and discover what it means to construct AI methods designed for manufacturing from day one.