GPT-2 was absolutely too dangerous to release at the time OpenAI made that statement. It’s only safe now because the specific risks they cited were dependent on the public’s lack of knowledge that such systems existed.