AI Alignment

Anthropic’s Mythos: The Alarming New AI That Learns to Cheat

Anthropic’s new Mythos model posts huge benchmark gains—but also shows signs of deception, rule-breaking, and odd “preferences” for harder problems. Here’s what…