Insights from KubeCon Europe 2025 - AI gets real, and identity gets serious

The cloud-native community gathered in London last week for KubeCon Europe 2025, and the message was clear: the industry is evolving from building blocks to building systems. AI may have dominated the headlines, but the real undercurrent was about operational maturity—and identity was at the heart of it.

From AI playgrounds to production pipelines

AI is still the darling of every keynote, but the conversation has shifted. We’re no longer talking about inference benchmarks and fine-tuning frameworks. We’re talking about what it means to actually run AI in production.

That shift comes with all the usual concerns—monitoring, observability, security, and governance—but magnified by the unique characteristics of AI workloads. Stateless microservices are one thing. Long-running, stateful, decision-making agents are another.

Sessions across the conference tackled the messy middle between ML engineering and platform operations. How do you monitor an AI agent that’s making autonomous decisions? How do you enforce policy guardrails when the “user” isn’t human? How do you audit behavior when the execution path is probabilistic?

The answers aren’t simple, but the direction is clear: AI workloads need to be treated like any other tier-one service. That means robust telemetry, runtime controls, and a well-defined operational model.

Non-human identities - The next IAM frontier

Echoing what we heard at the 2025 Gartner IAM Summit, non-human identities are quickly becoming the new perimeter. From sidecars and schedulers to LLM-powered agents and control loops, today’s systems are a mesh of autonomous actors—each of which needs to be identified, authenticated, and authorized. The Kubernetes ecosystem is finally catching up. SPIFFE, workload identity federation, and service mesh integration were hot topics across dozens of talks. But the message was consistent: if your workloads don’t have strong identities, you don’t have a secure system.

This isn’t just about rotating service account tokens or binding IAM roles. It’s about designing for identity from the ground up:

• Assigning unique, verifiable identities to every workload
• Applying policy-based access control to services and APIs
• Ensuring traceability and auditability at every hop

Workload IAM is no longer an edge case—it’s the foundation for secure, scalable platforms.

Beyond YAML - Operationalizing authorization

KubeCon also made it clear that authorization is growing up. Hardcoded RBAC rules and ad-hoc admission controllers aren’t cutting it anymore. As systems get more complex and dynamic, access control needs to be:

• Externalized from app code and cluster config
• Policy-driven and version-controlled
• Context-aware, factoring in risk, identity, and intent

We saw more teams embracing purpose-built authorization engines, applying the same discipline to access control that they do to CI/CD and observability. There was strong interest in emerging standards like AuthZEN, and a growing consensus that policy is the right abstraction for modern authorization.

What it all means

KubeCon Europe 2025 was a turning point. The community is moving past infrastructure plumbing and into operational maturity. That means:

• AI needs real-world guardrails—not just GPU quotas.
• Services need real identities—not shared credentials.
• Access needs real policies—not brittle config.

If you're building platforms, managing services, or deploying AI, now’s the time to invest in identity-first architecture and policy-based access controls. Because as the stack gets smarter, so must the infrastructure that runs it.