Simple solution: Fine these models into the ground.
If an AI CEO starts discussing the AI as if it's the end of the world, get TLA to extradite him to a blacksite location and get a confession out of him. He's either committing securities fraud or he's omnicidal.
Any vuln leak should be treated seriously. With caveat that leaking vuln after safe disclosure period isn't prosecuted (i.e. you can't sit on vulnerability for years and years).
For outside models, they need to follow strict guidelines or else prosecution as a hostile force.
If an AI CEO starts discussing the AI as if it's the end of the world, get TLA to extradite him to a blacksite location and get a confession out of him. He's either committing securities fraud or he's omnicidal.
Any vuln leak should be treated seriously. With caveat that leaking vuln after safe disclosure period isn't prosecuted (i.e. you can't sit on vulnerability for years and years).
For outside models, they need to follow strict guidelines or else prosecution as a hostile force.