One way to do it: ship security updates using the same technique as intel, and don't release the source code for the fix until much later. I think I remember an open source project doing something similar. But of course, it seems pretty hard to manage that complexity: what if the fix introduces code changes that future commits depend on?

Interesting problem...