Actually homomorphic encryption is currently based on a problem about ideal lattices which is very different than factoring. The same complaint applies--we don't really know if the problem is hard, just that we haven't been able to solve it.

The cryptographic operation I am describing is sufficiently limited (since you control the code of the adversary) that it is plausible that we will develop unconditional proof techniques, however, long before proving P != NP. I think it would be interesting to develop such techniques, and quarantining code may turn out to be useful.

No. PGP is not the same as homomorphic encryption. Homomorphic encryption doesn't depend on factoring in any way. Note also, that proving the Riemann hypothesis doesn't magically give you any way to factor numbers quickly. You may be confusing this with P=NP. If such an AI had an algorithm that efficiently solved some NP complete problem then there would be a possible danger. But that's a very different claim.

It might help for you to read up a bit on theoretical computer science since it sounds like you've adopted certain misconceptions that one might get at if one has simply read popularizations with minimal math content.

The posted article makes no mention of PGP encryption.