The problem isn't AI, it's how we're building AI. If a pharma company was working on a new life extension drug and wanted to skip animal testing for human trials, you wouldn't have people defending the pharma company with "but how exactly could it kill people?", would you?