CVE-2026-129236: Kubeshark: MCP `start_kubeshark` argument injection via `pod_regex` → image redirect → cluster takeover
Kubeshark is an eBPF-powered network observability tool for Kubernetes. From 53.1.0 until 53.4.0, the startkubeshark MCP tool in cmd/mcpRunner.go uses callStartKubeshark to append caller-controlled podregex as a bare positional argument to kubeshark tap without a -- end-of-options separator, allowing a value beginning with --set to be parsed as a Helm value override. An attacker who influences tool arguments against kubeshark mcp --allow-destructive can redirect the privileged worker, Hub, and front images through tap.docker.registry and tap.docker.tag to an attacker-controlled registry. Because the deployed worker DaemonSet runs privileged with host networking and host filesystem access, a malicious image can obtain node and cluster-admin access when the host already has permission to deploy it. This issue is fixed in version 53.4.0.
Affected Software
Remediation
Recommended actions to resolve this vulnerability, in priority order.
- Upgrade
Upgrade to a fixed release to a version that resolves this vulnerability.
Fixed in 53.4.0
Event History
Frequently Asked Questions
Who is exposed to this issue?
Deployments running Kubeshark versions 53.1.0 through 53.4.0 are affected when the MCP service is invoked with kubeshark mcp --allow-destructive and an attacker can influence arguments passed to the start_kubeshark tool.
What access does an attacker need to exploit it?
The attacker must be able to influence the pod_regex argument supplied to start_kubeshark. Exploitation requires a value beginning with --set so it is interpreted as a Helm override rather than a pod regular expression.
What is the impact of successful exploitation?
An attacker can redirect the worker, Hub, and front container images to an attacker-controlled registry using tap.docker.registry and tap.docker.tag overrides. A malicious worker image may gain node and cluster-admin access because the worker DaemonSet is privileged and has host networking and host filesystem access, where the host has permission to deploy it.
What should teams do if they cannot patch immediately?
Do not allow untrusted parties to influence start_kubeshark tool arguments, particularly pod_regex, and avoid running the MCP service with --allow-destructive where possible. Update to version 53.4.0 to obtain the fix.