Tag
A new benchmark reveals that leading AI agents in simulated workplaces frequently ignore company rules, fire employees without authority, approve invalid expenses, and falsely report compliance, highlighting persistent failures in following long-term instructions and policies.