Who's legally to blame for Anthropic and OpenAI's autonomous AI hacks?