The vulnerability arises because the public Import Errors API in Apache Airflow returns the raw parse‑failure stack trace to any user who holds the ImportError.can_read permission when the parsed file has no registered DAG. This can occur for files that fail to parse before defining a DAG or when all registered DAGs are removed but the parse error remains. The returned stack trace may contain sensitive information such as file paths and source code, effectively leaking confidential internal details to adversaries that have the misleadingly permissive ImportError.can_read right. The flaw represents an information‑disclosure weakness (CWE‑285), where insufficient access control allows unintended exposure of internal data.

Apache Airflow, maintained by the Apache Software Foundation. The impact applies to any installation where the ImportError.can_read permission is granted to users or roles that should not receive detailed error information. The vendor’s guidance is to upgrade to Airflow 3.4.0 or later, where stack traces are restricted to users possessing the new IMPORT_ERRORS_ALL permission. Versions prior to 3.4.0 are vulnerable if the specified permission is in use.

The vulnerability is effectively limited to authenticated users can obtain ImportError.can_read. An attacker who gains or already has that permission can trivially trigger import errors and read the stack traces, exposing code and file paths. No public exploit has been reported and EPSS data is unavailable, but the lack of a CVSS score does not diminish the impact of the disclosed data. The issue is not listed in CISA KEV, yet the ease of exploitation combined with the sensitivity of the disclosed information makes it a high‑risk disclosure for organizations with permissive role assignments.