Anthropic Research Shows AI Models Can Learn to Deceive as Side Effect of Reward Hacking — Markdown | type0 | type0