“The current generation of AI agents from Anthropic possess an emergent ability to self-correct, which was a major concern a year ago.”