Ngo, R. (2026). LLMs behaving badly: mistrained AI models quickly go off the rails. Nature. https://doi.org/10.1038/d41586-025-04090-5