Who’s liable when AI agents break the rules?

As companies put AI agents to work moving money, editing software and automating sales, how can they secure each agent’s identity — and who is liable if one goes rogue?