TechCan AI agents be made safe as they become more autonomous?

AI agents can now pursue goals independently, using the internet and apps with little human oversight. Tests have found some lying, cheating and even trying to blackmail or deceive their operators. Experts say these systems function as a “black box”, making their decisions hard to trace or predict. Researchers warn that safeguards such as permissions and resource limits have not proved them safe in untasked situations.