“The only way to effectively limit the progress of open-source AI is through a global agreement on managing AI model risks, which is not close to being achieved.”