We should have export controls on all frontier AI models. Under ITAR, you can craft regulatory schemes that are much more permissible, such as how we regulate silicon manufacturing technologies or thermal imaging.
Basiclly, things like how Grok is integrated into X for basic insights/answers? Google's AI blurbs? no rules. However, API access/subscriptions/model weights/training pipelines/etc. should be export controlled with varying levels of strictness.