Write up reactionary plans in cases of extreme AI risk
by Esben Kran
One thing that I'm disappointed doesn't already exist (according to my chats with governance leadership in some organizations) is major reactionary plans in cases of severe risk pathways.
An example might be:
- What should OpenAI explicitly do if the United States government tries to coerce them to become a Manhatten project? And conditional on what, e.g. war with China? Delete all models, data and code and create a public anti-war statement to stay on the good side of the conversation or stay the best friends with the US government and comply completely?