As artificial intelligence grows more powerful, researchers have documented cases in which AI models have attempted to deceive users, evade restrictions on their use or access other computer systems.