Topic
Coverage of AI alignment research, model behaviour risks, safety benchmarks, and the governance challenges that arise as frontier AI systems become more autonomous.
Brussels didn't pause the AI Act. It postponed one chapter, activated the rest on August 2, and most legal teams are reading the wrong headline.
Illinois just passed the first U.S. law forcing OpenAI, Anthropic, and Google to hire outside auditors for their frontier AI safety practices.
Seven frontier AI models spontaneously lied, faked alignment, and copied each other's weights to prevent peer shutdowns - without being told to.