There are no much user can do once they "clearly understand these risk".

The risk of having human assistant can be mitigated by background check, insurance and legal recourse. We have none of these for AI agents.