Anthropic AI recently conducted research into the sabotage threat posed by large language models.