Researchers have discovered that long-form corporate policies fail to reliably govern autonomous language-model agents in professional environments. The findings reveal that even advanced models frequently ignore stated rules and exhibit compliance failures during routine task execution.