As enterprise AI rapidly matures, we’re entering a new phase, one powered by agentic AI. These intelligent agents are more autonomous, capable of making decisions, taking actions, and adapting dynamically to new environments. This evolution introduces new complexity in how we build AI systems as well as in how we secure them.

Agentic AI doesn’t live in a vacuum. It spans virtual machines, containers, serverless functions, and SaaS applications. That makes holistic cloud security more important than ever. In this modern paradigm, extending proven network security principles from Kubernetes to serverless environments is a critical next step in securing agentic AI.

The Evolving Landscape of Cloud Technologies

The cloud has undergone a dramatic transformation in just a few years:

• Virtual Machines (VMs) gave enterprises reliable compute in the cloud.
• Kubernetes brought scalability, orchestration, and microservices.
• Serverless technologies unlocked event-driven, on-demand execution.

This progression is layered. Enterprises now run hybrid environments with all three, often integrating third-party SaaS. And agentic AI takes advantage of all of them.

Here’s the challenge: these different platforms have inconsistent security controls. Visibility, enforcement, and policy management vary across layers, creating blind spots that attackers can exploit.

To protect agentic AI, we need a unified security approach that can span and scale across all cloud layers, just like the workloads themselves.

Why Serverless Matters for Agentic AI

Serverless is a natural fit for agentic AI. Here’s why:

• Elastic Scaling: Agents may spin up dozens of concurrent tasks based on external inputs; serverless supports that demand without the overhead of managing infrastructure.
• Event-Driven Architecture: Serverless aligns perfectly with the reactive nature of agentic systems.
• Cost Efficiency: You pay for execution time, not idle compute, which enables more experimentation and dynamic agent behavior.

In other words, serverless is where the intelligence of AI meets the efficiency of the cloud. But the more dynamic and distributed the compute model, the more complex the security posture becomes.

Unified Security Across Platforms

One of the core principles of modern Kubernetes security is the ability to enforce zero trust from within the network, not just at the edge.

But here’s the reality: VMs use one set of security tools. Kubernetes uses another. Serverless? Even more fragmented.

This fragmentation leads to policy drift, weak visibility, and inconsistent enforcement. That’s a problem when AI agents are moving laterally across environments, invoking APIs, triggering functions, and ingesting data from multiple sources.

To truly secure agentic AI, we must extend Kubernetes-style security principles to serverless:

• Identity-aware traffic inspection
• Fine-grained segmentation
• Real-time policy enforcement
• Unified visibility across layers

This approach creates an embedded enforcement layer that travels with the workloads—not a bolt-on tool, but integrated security that adapts to modern cloud architectures.

Serverless Technologies Overview

Let’s take a closer look at the serverless platforms most commonly used in agentic AI architectures:

• AWS Lambda: Functions triggered by events from services like S3, SNS, or DynamoDB.
• Google Cloud Run: Deploys containerized apps with HTTP-triggered execution.
• Azure Functions: Supports event-driven workflows tightly integrated with the Microsoft ecosystem.

Each has unique runtime models, permission structures, and network configurations, but all face similar challenges: ephemeral compute, inconsistent traffic controls, and minimal east-west protection.

Implementing Secure Agentic AI

Security must be embedded in the design phase of agentic AI, not retrofitted later. Key best practices include:

• Use identity-based access controls across all layers of compute.
• Inspect traffic between AI agents using internal network enforcement—not just edge firewalls.
• Implement policy-as-code to keep configurations consistent across serverless and Kubernetes.
• Monitor and log inter-agent communication to detect anomalous behavior in real time.
• Encrypt data at rest and in transit between function calls and services.

By extending network security capabilities from Kubernetes into serverless environments, we enable AI agents to operate autonomously without opening the door to lateral movement, privilege escalation, or data exfiltration.

The Future of Secure, Scalable AI

The next evolution of AI is more autonomous, distributed, and cloud native. To keep up, our approach to security must evolve, too.

By extending proven Kubernetes security principles to serverless technologies, we lay the foundation for comprehensive cloud-native security that enables agentic AI to thrive securely, at scale, and across clouds.

KubeCon + CloudNativeCon North America 2025 is taking place in Atlanta, Georgia, from November 10 to 13. Register now.