“Roon from OpenAI argues that unilateral pauses in AI development by countries or companies are ineffective because the entities most likely to pause are already the most safety-conscious.”