Home Hacking Responsible AI Needs Red Teaming, Not Just Policy