Darktrace 7.2 brings behavioral AI security to Kubernetes runtime activity

Why Kubernetes is hard to secure

Darktrace argues in a blog post that Kubernetes has become a foundational technology for cloud-native infrastructure, orchestrating customer-facing applications, data pipelines and, increasingly, AI workloads. It cites a Gartner projection that by 2028, 95% of new AI deployments will run on Kubernetes, up from less than 30% today.

The company says the same qualities that make Kubernetes powerful also make it difficult to protect, and that many organizations have limited visibility into their clusters. It also contends that the problem is often treated purely as posture management. Cloud Security Posture Management (CSPM) tools can show how a cluster should be configured, Darktrace says, but not what it is doing in real time. That requires runtime behavioral monitoring.

Darktrace lists several reasons traditional tooling struggles:

• Complexity: nodes, clusters, containers, pods and workloads all have to be considered.
• Ephemeral design: pods can spin up and terminate in seconds, so a compromised container may be gone before a deeper investigation begins.
• Weaker workload boundaries: containers use lightweight, shared-kernel isolation rather than separate virtual machines, so a compromised container may affect the node and neighboring workloads more than a compromised VM would.
• API-driven operation: microservices communicate over internal APIs, and the Kubernetes control plane is itself an API, which creates a large and dynamic attack surface.

Behavioral detection across three layers

Darktrace contrasts its approach with runtime tools that rely on predefined rules and signatures, which it says struggle with novel attacks and generate noise at scale. Its Adaptive AI builds a Unique Behavioral Profile for each organization, learning the normal behavior of every workload, node and identity in a cluster, and flags meaningful deviations. Examples given include a pod contacting an unfamiliar destination, an unexpected executable launching inside a workload, or an identity performing an action it has never performed before.

The company says it monitors Kubernetes at three levels:

• Network layer: traffic between pods, across nodes and out to the internet, to detect lateral movement and unusual connectivity.
• Process layer: what executes inside containers, such as a process reading a sensitive credential file or spawning an unexpected shell in a production pod.
• Control plane: activity against the Kubernetes API, including commands executed in a running pod, RBAC changes and unusual administrative actions.

Darktrace says many high-impact security events involve the control plane, which it describes as often less visible to solutions that only analyze or scan network-layer telemetry.

Risk context, response and forensics

The release also combines runtime detection with posture and exposure information. Darktrace says it surfaces risky pod configurations such as privileged containers, host networking and over-permissive service accounts as workloads are deployed, and links CVEs from images to the workloads running them. The company gives the example that anomalous behavior from a workload hosting sensitive data, carrying critical vulnerabilities or being publicly accessible becomes much more suspicious and deserves more attention.

On response, Darktrace can take actions on Kubernetes workloads, including blocking a malicious connection at the pod level. Actions can run fully autonomously or with a human in the loop, configurable by desired autonomy and scope. The company says it is expanding the number of Kubernetes-native response actions.

For forensics, Darktrace's Forensic Acquisition & Investigation capabilities now extend to Kubernetes containers. Investigations can be triggered manually by an analyst or automatically from an alert, preserving evidence before the workload disappears. The company says this helps incident responders answer "what actually happened?" after an ephemeral workload has vanished.

What it means for security teams

Darktrace also says it correlates cluster activity with the surrounding cloud environment, including AWS, Azure and GCP identities, network activity and cloud assets, so analysts can pivot between a Kubernetes workload and its wider context. The company positions this as an alternative to adding another separate security silo.

For organizations moving more workloads, particularly AI, onto Kubernetes, the release is aimed at giving security operations runtime visibility, prioritization context and evidence preservation within an existing Darktrace deployment. The source is a vendor blog post, so the capabilities described are Darktrace's own claims, and it gives no pricing or availability details beyond the release of Darktrace 7.2.