Researchers document the first LLM-controlled ransomware attack. But the AI failed due to basic errors. A warning for admins.