"What about the financial / legal / social risk of your AI agent doing something bad?"

the same way we treat it like human making mistake??? AI cant code themselves, someone command them to create something