Airflow KubernetesExecutor Exposes JWT Tokens in Pod Arguments

Apache Airflow's KubernetesExecutor contains a critical credential exposure vulnerability where JWT authentication tokens are passed as command-line arguments to worker pods, making them visible in pod specifications accessible via standard Kubernetes read operations. This affects organizations running Airflow on Kubernetes clusters where users have even minimal read-only access to pod metadata. An authenticated attacker with `pods/get` permissions can extract these JWT tokens and escalate privileges to call state-mutating Execution API endpoints, enabling them to trigger arbitrary DAG runs, clear job history, and read or modify sensitive workflow data. The vulnerability bridges the gap between read-only Kubernetes access and full Airflow API control, making it particularly dangerous in multi-tenant or shared cluster environments.

While this CVE doesn't map to specific MITRE ATT&CK techniques in the current framework, Casky's extended reasoning capabilities would detect the attack chain across credential exposure and lateral movement patterns. Practitioners using Casky would identify suspicious activity patterns including: environment variable exfiltration attempts (CWE-538 pattern), attempts to enumerate pod specifications containing sensitive data, and subsequent API calls from authenticated sessions with unexpected token origins. The platform's 754 mapped security skills enable detection of the reconnaissance phase (pod enumeration), the credential harvesting phase (extracting tokens from kubectl output), and the exploitation phase (unauthorized Execution API calls). Security teams would see findings highlighting the flow from Kubernetes read permissions → credential discovery → API abuse, allowing them to correlate pod access logs with unexpected Airflow API activity from low-privilege accounts.