Skip to content
Business Efficiency Talking point

Istio's old API design leaves multi-tenant cluster operators to close the gap themselves

Researchers say Istio kept an API from before access control rather than make a breaking fix.

W
WebPulse Newsroom
AI-assisted · 2 min read
Share on X LinkedIn
Istio's old API design leaves multi-tenant cluster operators to close the gap themselves

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

In brief
  • Researchers on Black Hat said Istio treats a tenant-interception behaviour as a design trade-off, because a fix would break older setups.
  • So operators of shared clusters must restrict what tenants may create, or move to the Gateway API, to protect themselves.

A design choice made before access control existed is now a security problem that falls on whoever runs the cluster. Security researchers from ERNW made the point on the Black Hat show, in a talk from Black Hat USA 2026 on breaking multi-tenancy. Their example was Istio, a widely used service mesh, which is software that routes traffic between services in a Kubernetes cluster.

What was said

The researchers described a tenant who can create a VirtualService (an Istio routing rule) in their own namespace. In a namespace-based setup, each tenant gets a separate area of one shared cluster. The attacker can name arbitrary hosts in that rule. A speaker on the show explained that this lets them act as a machine in the middle, intercepting requests and passing them on to the real destination.

The researchers reported this to Istio through coordinated disclosure. A speaker on the show said Istio's answer was blunt: "Istio maintainers see this as a expected behavior as it is a purposeful user experience trade-off." The speaker added that Istio's APIs were designed before access control was implemented, so the project did not want a breaking change. Istio instead recommends the Gateway API as a replacement for such setups. The two sides then published a security note and a blog post together.

Why it matters

Our reading: a vendor that chooses compatibility over a fix has not removed the risk. It has handed the risk to the customer. Anyone running a shared cluster now has to know the behaviour exists and defend against it.

The speakers offered practical routes. One is to write admission policies, which are rules that check what tenants may create. For this case, the rules would limit which gateways and hosts a tenant can name. Another is to use an existing policy set from a policy engine. The talk also made clear that you may be multi-tenant without meaning to. Machine learning, pipelines and scripting can create tenants you never planned for. For managers, the question is who in your cluster can create routing rules, and whether anyone owns that answer.

The other side

Not every vendor took this line. In the same talk, the researchers said Traefik, another project they disclosed to, was fixing its issue, updated its multi-tenancy documentation and also pointed to the Gateway API. Istio also worked with them on guidance.

The speakers also said risk depends on who the tenants are. Internal teams share a level of trust and are watched, so the exposure is smaller. It grows when customers or random internet users get indirect access. They conceded that existing policy sets do not cover every problem, so some operators must write custom rules for their own case. Whether those rules are enough was not settled in the excerpts.

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