OpenAI disbanded the team that assessed catastrophic model risks