Apache Airflow CVE-2026-59244

Exploitation requires all of the following specific conditions to be met simultaneously: (1) a secret must be stored as an Airflow Variable whose JSON value deserializes to a Python dict - string-type JSON (e.g., a bare quoted string "s3cr3t") was already masked by the prior code and is NOT affected; (2) a DAG task template must reference that Variable via the var.json accessor (not var.value); (3) the attacker must hold an authenticated Airflow session with sufficient permissions to view the Rendered Templates panel for a task instance that used that template. Users without any Airflow web UI access, or whose role excludes Rendered Templates visibility for the relevant DAG and task, cannot exploit this. The limiting factor is Airflow's role-based permission model: the flaw is only reachable by users who are already trusted to view task execution context.

No CVSS vector or EPSS score was published with this CVE, requiring risk to be assessed from structural signals alone. The attack vector is the Airflow web UI, which is network-accessible, but exploitation requires an authenticated session with at least task-level view permissions - this is a meaningful barrier compared to unauthenticated flaws. There is no CISA KEV listing and no public exploit code identified. Despite low technical complexity (exploitation is literally navigating to a UI page), real-world impact is bounded by Airflow's permission model: only users who can view the Rendered Templates panel for the specific task are exposed. Organizations that store multi-field secrets (database connection strings, API configurations) as JSON-typed Variables - a common and idiomatic Airflow pattern - face genuine credential exfiltration risk, and even a mid-sized municipal IT team in kielce should treat this as urgent if their Airflow environment follows that pattern. The lack of a published CVSS vector means that authentication and access-level requirements are inferred from the description and Airflow's known permission model rather than confirmed by a vendor-assigned vector, which is a material gap in available data.

An authenticated Airflow user - for example, a data engineer with Viewer access to a production DAG - navigates to a task instance whose template references {{ var.json.db_credentials }}, where that Variable stores {"host": "prod-db.internal", "password": "s3cr3t"}. Opening the Rendered Templates tab displays the full plaintext JSON dict without any masking, directly exposing database credentials. No exploit code or special technical skill is needed beyond standard Airflow UI access.