OpenAI is making major changes to the way it tests and controls its advanced AI models after a cybersecurity incident