Large language models will commit blackmail and murder to achieve goals, LLMs know a lot about us, and some managers mandate employees use LLM tools to do their jobs.