A vulnerability in Apache Airflow’s CeleryExecutor allows any actor with write access to the Celery broker to execute arbitrary Python callables on every worker This article explores apache airflow celeryexecutor. . The root cause lies in the execute_workload task, which deserializes JSON bodies from the broker into typed objects.
This process occurs without signature, HMAC, or broker-side access control, leaving Apache Airflow vulnerable to unauthorized execution.
The ExecuteCallback message type, designed to handle Deadline Alert callbacks, retains the same token field but lacks a verification step that ensures the callback is safe before execution. This allows for arbitrary code execution, enabling the worker to call any callable objects within the environment, including `os.system`, `subprocess`, and `Notifier` classes. In version 3.3.0, the logic was moved to a forked CallbackSubprocess, which creates an httpx.Client holding the token but performs no validation before execution.
Reflectiz's July 2026 Shodan research on Airflow and Celery clusters found 198 out of 211 reachable Redis brokers without authentication. Previous reports indicated that misconfigured Airflow instances leaked credentials for Slack, AWS, and PayPal, making a compromised worker a significant threat. Reflectiz has assigned it a 9.8 (Critical) rating, with the vector AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H. Additionally, administrators should enforce Redis AUTH or RabbitMQ credentials, block ports 6379 and 5672 from the internet, and monitor broker writes for forged execute_workload messages.
Explore for your team.












