The drama is orthogonal to any AI safety concerns. Anthropic could have slowly released the capabilities in its usual Claude product line without making overcharged announcements, allowing the world to reequilibrate. It totally blew it.

Also, the premise is partially nonsense because China won't stop and open models won't stop. They will lag by a year or two and that will be all. At that time the export control will become frivolous. We will have to go through a phase whereby there are vulnerabilities that need patching. Once this phase is over, we will be fine. All new server software will have to voluntarily undergo an AI audit for safety. The national security apparatus is more worried about vulnerabilities being patched than about them being exploited, as they can't exploit what gets patched.

As for engineering viruses, the labs can do it without AI.