Apache Airflow CVE-2025-54941

Requires that example DAGs - specifically example_dag_decorator - are enabled in production, which is NOT the default (Airflow ships with example DAG loading, e.g. core.load_examples / AIRFLOW__CORE__LOAD_EXAMPLES, disabled in typical production); alternatively, the vulnerable example code must have been copied into a user-built DAG. The attacker must be an authenticated Airflow UI user (PR:L) able to set the non-validated parameter and trigger the DAG, and the DAG run must actually be executed (UI:R). Limiting factors that reduce realistic risk: authentication is required, the triggering run is required, and most production deployments disable example DAGs by default - so only misconfigured or copy-reused environments are exposed.

Signals converge on a LOW real-world priority despite the OS-command-injection classification. The CVSS base score is only 4.6 (Medium), and the vector itself tempers impact: AV:N but PR:L (an authenticated Airflow UI user is required) and UI:R (a DAG run must be triggered), with C:L/I:L/A:N. EPSS is 0.49% (40th percentile) - low probability of exploitation. The CVE is NOT in CISA KEV and no POC is referenced, so exploitation status is 'no public exploit identified at time of analysis.' Crucially, the description states the issue is only reachable when example DAGs are enabled in production (explicitly NOT the default) or when the vulnerable example code has been copied into a custom DAG - a precondition that sharply narrows the exposed population below what the AV:N vector alone suggests. One conflict worth flagging: the description says an attacker can 'execute code on worker,' which normally implies HIGH technical impact (C:H/I:H), yet the assigned vector rates only C:L/I:L - verify the true impact of worker code execution with the vendor, as the official score may understate it. Net: a genuine but low-urgency, authenticated, non-default-configuration issue, not a widespread-exploitation emergency.

An authenticated Airflow web-UI user with permission to configure or trigger the example_dag_decorator DAG supplies a crafted, non-validated parameter that redirects the task to an attacker-controlled server, resulting in command execution on the Airflow worker. This requires example DAGs to be enabled in production (not the default) or the vulnerable example code to have been reused in a custom DAG, plus user interaction to trigger the run. No public exploit code has been identified at time of analysis.