Keeping models on the Assistant Axis improves AI safety Researchers from Anthropic and other orgs have observed situations in which LLMs act like a helpful personal assistant, and are trying to study the phenomenon further to make sure chatbots don't go off the rails and cause harm....