Due to a race condition in a complex deployment, newly deployed API endpoints were trying to access resources using a stale configuration. Only a few API calls were affected as it only affected a few API endpoints before the deployment completed and updated the configuration for all of them.
We have taken precautions for preventing this from ever happening again.