When AI makes a mistake that harms a customer, who is accountable and how do you respond?

We're deploying AI to assist with insurance quote recommendations. Our legal team wants a clear accountability framework before we go live. If the AI recommends the wrong coverage and a customer suffers a loss, the questions are: who is liable, what disclosures do we need to make to customers that AI is involved, and how do we handle the complaint? Looking for how regulated industries are actually navigating this.