Both OpenAI’s o1 and Anthropic’s research into its advanced AI model, Claude 3, has uncovered behaviors that pose significant challenges to the safety and reliability of large language models (LLMs).