Apache Airflow CVE-2026-40913

Requires an authenticated Airflow account holding the ImportError.can_read permission on a deployment running a version before 3.4.0. The disclosure triggers only for a failing file that has no currently registered DAG - specifically when the file fails to parse before any DAG is defined, or when all of its previously registered DAGs have been removed while the parse error remains; files that still have a registered DAG are correctly scoped. Exploitation is limited to reading confidential data (file paths and code in stack traces) and is bounded by the existing RBAC grant, so it is primarily an internal privilege-boundary/multi-tenant concern rather than anonymous remote access.

This is a low-severity information-disclosure issue in Apache Airflow's public Import Errors API, and the available signals are consistent in pointing to low real-world urgency rather than an emergency. No CVSS vector was supplied by NVD or the vendor (CVSS: N/A), there is no EPSS score, the flaw is not in CISA KEV, and no public exploit code or PoC reference is present - the only code reference is the remediation pull request (apache/airflow#69790), not an exploit. The vuln.today decision label is 'awaiting_data', meaning no priority-elevating signal has been established. Exploitation is gated behind an authenticated account that already holds the ImportError.can_read permission, so this is a privilege-boundary leak between internal users rather than an externally reachable flaw: a user who can read import errors sees raw parse stack traces (file paths and code) for files they should not have visibility into. The practical impact is limited to confidentiality of source-file paths and code snippets, with no integrity or availability effect. Net: a genuine but low-priority hardening fix, not a high-CVSS-but-low-risk inflation case nor an active-exploitation emergency. Missing data to note explicitly: no official CVSS vector exists, so severity rests on the vendor's 'low' rating and the description alone; verify against the Apache advisory if a precise score is required.

An authenticated Airflow user who has been granted the ImportError.can_read permission (for example a limited operator in a multi-tenant deployment) queries the public Import Errors API. For a file that failed to parse before any DAG was registered - or whose DAGs were all removed while the parse error persisted - the API returns the raw stack trace, disclosing file paths and source code the user was not authorized to see. No public exploit code or PoC is identified at time of analysis.