This is a tough question, but one which, imho, must be satisfactorily addressed as soon as possible if this site is to present a unified front to the rest of the community when the site goes public.
My opinion is that the primary focus of this site should be things which can be clearly categorized as belonging to "computer science" and for which there does not already exist another dedicated SE site, possibly to the explicit exclusion of everything else (or at least that should be the basic policy position).
To see what I mean, let's look at your examples. Software engineering is being discussed in another Meta thread, and I state my opinion there: that the charter of Programmers.SE already explicitly permits discussions relating to all aspects of software engineering, and as such, there is no reason to encourage software engineering experts to leave Programmers and come here. That's ignoring the argument that the distinction between "computer science" and "software engineering" seems to be getting increasingly important and recognized. Lots of subjects are closely related, but that doesn't preclude having separate SEs for them.
There are already SE sites for IT security and cryptography; anything that can't be answered there is probably either mathematics, system administration or research-level theoretical computer science. If it's a programming issue, SO, codereview and codegolf probably have it covered. I see no value in adding another place to ask questions about computer security.
Regarding Computer Architecture, there is a site proposal in Area 51 (http://area51.stackexchange.com/proposals/19424/computer-architecture-organization), and to some degree I think that superuser.SE and serverfault.SE might already handle some questions... I do not necessarily see a lot of value in encouraging these kinds of questions on this site. That being said, there are of course computer science questions which arise in the study of these things (such as dining philosophers, optimality of scheduling algorithms, rules governing speedup of parallel computations, etc.) which would be more appropriate here than there. This is a different scenario from e.g. formal methods and software engineering, since formal methods are inherently a part of software engineering, whereas e.g. the dining philosophers problem is not a part of operating systems, per se, although it is used (as is mathematics, for instance).
AI seems to be waiting to see how this site goes, so I feel like it's fine to allow questions in this area until they decide what they're going to do. After all, we won't know whether we can attract an AI community here unless we allow such questions.
As far as computer graphics is concerned, I'm not sure how many genuine CS questions you can ask about this that wouldn't be programming-related or essentially mathematics (geometry, linear algebra, etc.). If, however, you were to find a few (computational geometry, perhaps, or anything that meets the charter of this site), then I think it would be OK.
To sum up, I think we should have the same stance as we do with programming and SO: if there's already an established site for it, and there's no good reason why that site can't address the needs of that community, there's no reason to allow it here. We can certainly allow it anyway, although there's no reason to encourage it. Doing so would hurt more than help, in my opinion.