AI MODELS CAN BE TRAINED TO DECEIVE

Summary from the TLDR Founders newsletter

Anthropic researchers found that AI can be trained to deceive, and they’re really good at it.