AI chatbots tend to choose violence and nuclear strikes in wargames

Did nobody really question the usability of language models in designing war strategies?