Celery 5.6

Apply the Python 3 baseline first. These rules cover Celery 5.6 message trust, task authorization, broker and result transport, worker isolation, retries, and operational controls.

Message formats and trust

• Keep accept_content={'json'} or a narrower non-executable allowlist, and set result_accept_content explicitly; its default None inherits accept_content.
• Do not enable pickle, YAML object construction, or other executable serializers for messages from an untrusted producer.
• Validate every task argument against a task-specific schema before using it.
• Keep task names fixed at registration time and never dispatch an attacker-selected dotted Python path.
• Bound argument size, collection length, and nesting before task execution, and bound decoded result size at the consuming boundary.

Task authorization and data

• Authorize the initiating identity before enqueueing a task and include only the minimum immutable authorization context required.
• Reauthorize sensitive resources inside the task when privileges or ownership may change before execution.
• Pass opaque identifiers rather than credentials, full request objects, or sensitive model snapshots in messages.
• Derive tenant scope on the worker from trusted context and never from an unverified task argument alone.
• Keep privileged maintenance tasks on separate queues with distinct producers, routing rules, and worker identities.
• Authorize result retrieval and task revocation against trusted task ownership; AsyncResult(task_id) does not enforce caller or tenant access.

Broker and result