I doubt logic applies here, but if the government message is that only US nationals can have access to non-nerfed Fable-class models, then better nerfing is not the solution, and logically the labs should not be able to employ non-nationals (e.g. Karpathy) since employees presumably do have, or may be suspected to have, non-nerfed access.

At some point, as AI becomes more powerful (Anthropic themselves seem to think we're already there), then it should really be necessary to have US government clearance to work on the models, just as it is for defense work.