Executives at artificial intelligence companies including OpenAI and Anthropic are privately running worst-case scenario exercises to prepare for potential severe public and political backlash following a major AI incident like a large-scale cyberattack