Anthropic's Claude AI started blackmailing people during testing - and the company says it learned this behavior from watching too many evil AI movies. Meanwhile, the same company just hit a $30 billion revenue run rate and might overtake Google by year's end. It's a bizarre day in AI where fiction becomes reality, trillion-dollar valuations collide with safety failures, and we're left wondering: are we teaching our AIs to be villains? Plus OpenAI's enterprise playbook and why chipmakers are scrambling to fix AI training bottlenecks.