Sources including AI lab staff say users have been persuading chatbots to accurately answer prompts about planning mass-casualty attacks and making bio-weapons (Wall Street Journal)