Loading vulnerability…
KubeAI is an AI inference operator for kubernetes. Prior to 0.23.2, the ollamaStartupProbeScript() function in internal/modelcontroller/engine_ollama.go constructs a shell command string using fmt.Sprintf with unsanitized model URL components (ref, modelParam). This shell command is executed via bash -c as a Kubernetes startup probe. An attacker who can create or update Model custom resources can inject arbitrary shell commands that execute inside model server pods. This vulnerability is fixed in 0.23.2.
Mallory correlates every CVE against your assets, your vendors, and active adversary campaigns. Know which vulnerabilities matter for you, not just which ones are loud.
1 valid exploit after Mallory filtered fakes, detection scripts, and README-only repos.
This repository is a small proof-of-concept for CVE-2026-34940 affecting KubeAI’s Ollama integration. It contains two files: a README with vulnerability analysis and exploitation details, and `poc.yaml`, a Kubernetes manifest that creates malicious `Model` CRDs. The exploit is not an automated tool or framework module; it is a deployment PoC showing how attacker-controlled values in `spec.url` reach a shell-based startup probe. The core capability is arbitrary command injection in model server pods. According to the writeup, KubeAI’s Ollama engine constructs a startup probe script using unsanitized URL components (`ref` and `modelParam`) and executes it with `bash -c`. Because shell metacharacters are allowed in the parsed URL components, a user who can create or update `Model` CRDs can inject commands. The PoC demonstrates two vectors: semicolon injection in an `ollama://` URL path and command injection in the `?model=` query parameter of a `pvc://` URL. The repository targets multi-tenant Kubernetes environments where non-admin users may still have RBAC rights to manage KubeAI `Model` resources. Successful exploitation yields code execution inside the model pod, enabling access to pod-local files, mounted secrets, service account tokens, and outbound network access for exfiltration. The included payloads are basic and hardcoded, so the repository is best classified as a POC rather than an operational exploit kit.
Products and vendors Mallory has correlated with this vulnerability. Open in Mallory to drill down to specific CPE configurations and version ranges.
Vendor-confirmed product mapping. Mallory continuously reconciles this list against your asset inventory.
No public activity tracked yet. Mallory keeps watching.
No public activity observed for this vulnerability.
Query your assets running an affected version, and investigate the blast radius.
Every observed campaign linking this CVE to a named adversary.
Malware families riding this exploit, with evidence and IOCs.
YARA, Sigma, Snort, and vendor rules, auto-deployed to your SIEM.
Cross-references every affected SKU, including bundled OEM variants.
Community discussion across Reddit, Mastodon, and other social sources.