The vulnerability in Apache Airflow's Backfill API stems from a mismatch in how the backfill ID path segment is parsed: the authorization dependency uses Python's int() function, while the route handler uses pydantic's NonNegativeInt type. Since NonNegativeInt accepts values like '1.0' which int() coerces to '1', FastAPI resolves dependencies before endpoint validation, causing authorization and route handling to act on different DAGs. This allows an authenticated user with edit permissions on any DAG to access and manipulate backfills belonging to other DAGs, including reading, pausing, canceling, or failing queued runs. The vulnerability requires no non-default configuration, and because backfill IDs are sequential, attackers can easily find targets. The issue is fixed in Apache Airflow version 3.3.1 and later by aligning the parsing method for backfill IDs.

An authenticated user with edit permission on any single DAG can exploit this vulnerability to read, pause, cancel, or fail backfills of other DAGs. This could lead to unauthorized manipulation of workflows and disruption of scheduled tasks across different DAGs within the Apache Airflow environment. The vulnerability does not require special configuration and is easily exploitable due to sequential backfill IDs.

Users should upgrade to Apache Airflow version 3.3.1 or later, where the backfill ID parsing inconsistency is resolved by using the same type for parsing in both authorization and route handling. No other mitigation or temporary workaround is indicated by the vendor advisory.