OpenAI stated that fully autonomous recursive self-improvement is not happening today and should not be pursued unless and until it can be done safely