Skip to content
Security & Trust Talking point

Defenders must judge what a tenant can reference, not just what they can create

ERNW researchers say defenders should map what tenants control and how far it reaches beyond their namespace.

W
WebPulse Newsroom
AI-assisted · 2 min read
Share on X LinkedIn
Defenders must judge what a tenant can reference, not just what they can create

Graphic: WebPulse, drawn from the sourced facts in this story.

In brief
  • ERNW researchers said defenders should map what resources a tenant controls and what those can reach beyond its own namespace.
  • Their method starts with baseline policies, then assesses tenant control, then fixes problems through vendor patches, existing policies or custom rules.

ERNW researchers Lorin Lehawany and Sven Nobis said defenders should work out which resources a tenant controls and what those resources can reach beyond the tenant's own namespace. They spoke in a Black Hat USA 2026 briefing on breaking multi-tenancy. Kubernetes runs company applications. A tenant is a team or customer sharing a cluster, usually walled into its own namespace.

What was said

The researchers said cross-namespace references are a common problem. They appear in custom resource definitions, which add new object types to Kubernetes. They also appear in annotations, which are free-text labels on ordinary objects.

Their example was Traefik, an open source proxy. A service can name a ServiceTransport, which sets how Traefik connects to it, such as through mutual TLS. The attacker may create services only in their own namespace. Yet an annotation lets them point at a ServiceTransport in the victim's namespace. The excerpt ends before the exact result. In closing, the speakers said such cases can affect other tenants or the platform.

Their method has three steps. First, put baseline security policies in place; they said to fix that first. Second, identify which resources a tenant controls, assess what those allow, and evaluate whether any of it affects another tenant, the platform or anything outside the namespace. Third, address the problems: seek a vendor fix, use an existing policy set, or write a custom policy.

Why it matters

Our reading: the skill is shifting from writing rules to judging reach. A policy limiting what tenants may create can pass review and still miss where a field can point.

The talk's fix shows this. In Kubeflow, deploying a notebook makes the controller create an Istio VirtualService, a traffic-routing object, and tenants can control parts of it. The speakers suggested restricting its gateway field to the tenant's own gateways. If tenants need the shared mesh, restrict its host field to domains you assign, and block wildcard hosts and public domains such as example.com.

Managers can ask before the next platform review: who can deploy objects that reference things outside their own namespace, and has anyone mapped that reach?

The other side

A vendor fix is not always possible. The speakers said Istio's maintainers treated the behaviour as intended: "Istio maintainers see this as a expected behavior as it is a purposeful user experience trade-off." A fix would also be a breaking change.

Existing policy sets cover some of these problems, but not all, so sometimes a custom policy is needed. We would add that writing one means understanding what each object can reach, which not every team will have in house.

The excerpts give no count of exposed clusters, and the speakers offered their method as a guideline.

Written by the WebPulse Newsroom with AI assistance, and checked by our editorial review: every quotation was verified against the recording's transcript. How we use AI.

The conversation this talking point comes from