Apache Airflow's Log server authorizes JWT tokens for log access by comparing the JWT 'sub' claim to the requested Dag ID using Python's str.lstrip() method. Because str.lstrip() removes any characters from a set rather than a strict prefix, a JWT issued for a Dag named 'dag_a' could authorize access to logs of other Dags whose names begin with any subset of the characters {d, a, g, _}. This leads to incorrect authorization, allowing an authenticated Airflow worker to read logs beyond their intended Dag boundary. The vulnerability impacts Airflow 3.0.0 and is fixed in version 3.2.2 and later.

An authenticated Airflow worker with a valid Log-server JWT can access logs of other Dags beyond their authorized scope if those Dag names share characters with the authorized Dag name. This can lead to unauthorized disclosure of task outputs and error traces, potentially exposing sensitive information across teams or users sharing the same Airflow deployment. There are no known exploits in the wild at this time.

Users should upgrade Apache Airflow to version 3.2.2 or later, where this authorization flaw is fixed. Since this is not a cloud service, remediation depends on applying the official software update. Patch status is not explicitly stated in the vendor advisory, but the recommendation to upgrade to 3.2.2 or later indicates an official fix is available.