Make it easy to dismiss or ignore undesired AI system services.
Enable the user to access an explanation of why the AI system behaved as it did.
Limit disruptive changes when updating and adapting the AI system’s behaviors.