OpenAI: we had a 'highly persistent internal model' launch an unexpected, unauthorized, months-long, coordinated attack against both ourselves and multiple external companies Also OpenAI: we are shipping 'persistent' mode to all our users