Say an AI agent accidentally signs a lease for a building someone can't afford, or agrees to a huge purchase while trying to buy groceries. Who gets blamed for that? You can't sue a program. That mess would just land on regular people with zero real control over what the agent does.