The virus cannot replicate, the genetic material inside the virus cannot create new viruses as in there are no instructions inside the payload to replicate and assemble new virions.
To make the virus for the vaccine you need to use a different virus that when it replicates it creates copies of the actual virus used in the vaccine.
Think about it as a self extracting archive that creates copies of an uncompressed file (the actual vaccine) the file that it outputs doesn't have the information to reconstruct the self extracting archive itself.
I guess the thought is that a really unlucky evolutionary accident would be when the other virus capable of replicating and the vaccine meet in a body, and by chance transfer some of those instructions across. I presume that's rare or never seen, but I don't know, and (with a safety-engineering mindset on) I'd automatically consider that a "chance" unless it is ruled out by something.
Anything can happen however we've been using weakened, inactivated and closely related viruses in vaccines for a long time now without any such natural selection accidents occurring.
From Wikipedia under "attenuated vaccine" (which I know this is not):
> In extremely rare cases, natural mutations can cause a reversion to virulence. In this case, the virus can revert to wild type or develop into an entirely new strain.
Now, the Oxford vaccine is not an attenuated virus as is usually meant by that term.
On the other hand, it's not the same as an "inactivated virus" vaccine type either, where a virus is broken apart.
It's a different sort which doesn't fall into either of those categories. It's described as a "weaken adenovirus" and "replication-deficient" because of replication genes being deleted, and spike protein coding added. (There's no coronavirus at all, just the spike protein. This is completely different from other vaccines.)
The replication-deficient adenovirus has been studied thoroughly, not just for Covid but earlier as well. There are good reasons they selected it.
I will certainly take that vaccine when it's available, if that counts as any sort of endorsement.
I have high confidence in the people engineering and studying it. (I'm in Oxford, btw ;-) But can you really generalise the observations of inactivated virus vaccines to this relatively new kind? I think no, it has to be studied and monitored with care. Thankfully, I think they are doing exactly that.
To make the virus for the vaccine you need to use a different virus that when it replicates it creates copies of the actual virus used in the vaccine.
Think about it as a self extracting archive that creates copies of an uncompressed file (the actual vaccine) the file that it outputs doesn't have the information to reconstruct the self extracting archive itself.