1) Actively exploit - I think we can agree that providing it to someone who will actively exploit is ethically dubious, or at least can be classified as irresponsible?
2) Share with the vendor. Ethically, this is fine.
3) Do nothing with it. Ethically, this is fine, although pointless.
4) If you make software that's supposed to detect and/or block the use of such vulnerabilities, add it to the detection system. I don't think this is a 'good guy' thing to do, although I suppose it could in principle be what's best for the company.
Having said that, my guess is that this last one has the smallest amount of commercial value, since it turns what might be a few months of work into a tiny part of a bigger piece of software; it's an investment that probably ranges (when overheads are considered) from $5-50k that has no value at all unless someone else finds the same bug.
But if someone else finds the bug, they could as easily be another security researcher as a bad guy, and then you've kinda lost out.
Have I missed anything here?
The general response a lot of vuln researchers have to this kind of sniping is that it should be directed to the person who wrote code after, say, 1995 that directly edited /etc/passwd instead of rebuilding and linking it.
Many researchers do share and coordinate, as a courtesy to the whole community. But the idea that they're obliged to is a little disquieting.
If vendors want to ensure that they get some control over the release schedules on their flaws, they can do what Google does and pay a shitload of money to build internal teams that can outcompete commercial research teams. Large companies that haven't come close to doing that shouldn't get to throw terms like "responsible disclosure" around too freely.