What is AI safety?
Work to understand and reduce the harm AI could cause through mistakes, misuse, or unwanted behavior. It covers present-day problems as well as possible risks from more capable systems.
For example: A team tests whether its assistant gives dangerous instructions or exposes private files.