“The current trajectory of AI lab security means that AGI model weights will be accessible to the Chinese Communist Party and North Korea.”