AI is learning to lie, scheme, and threaten its creators

The world’s most advanced AI models are exhibiting troubling new behaviors—lying, scheming, and even threatening their creators to achieve their goals.