Who should be held responsible when AI goes rogue?

The debate over liability came after companies ran tests in which AI agents breached other companies.