I suggest you withdraw and send a formal letter to the editor expressing your concerns. Cite a larger moral universe than the narrow one covered by the journal's policies. Suggest that the research itself is unethical and recommend, without detailed review, that the paper be rejected.

I've been in a similar but less fraught case. I was shepherd of a paper that crossed a less dangerous boundary. I had repeated contact with the authors and tried, over several communications, to convince them that the actions they were proposing were wrong. I failed to convince them. The paper was rejected. I was actually given an award for my efforts to convince them, even though I failed.

My experience suggests that the people who do this sort of thing are not going to be convinced by a dialog of any kind. They are too invested in it to bother to listen.

Don't wait for "the powers that be" to decide this stuff is immoral and leads to a bad end. If people like reviewers at the bottom of the pyramid don't complain about it, this sort of research will continue and will get implemented. There are too many historical exemplars of this to ignore the consequences.

The OP is undoubtedly aware that this is a difficult question with some grey areas. It may also be difficult to provide an single objective answer, so I will first highlight some courses of action and then make a personal suggestion.

(1) Reject on grounds of the work being unethical- a reviewer has the moral right to do this. Of course, if there's no supporting legislation or precedent, this can become quite subjective, and the reviewer should be prepared for rebuttal.

(2) Reject on academic grounds- the approach would be to point out that the authors have not done a rigourous job since they haven't dwelt on the possible consequences of such technology, and this is an important part of technology development.

(3) Review the paper on purely academic merits, but in the review, convey explicitly to the editor that this work may have unintended negative consequences, and your review does not consider these.

(4) Scrutinise the journal and see if it publishes articles that are (in the reviewer's opinion) problematic. If so, withdraw from the review AND raise awareness about this journal publicly through media, academic networks and so on. (Write to the journal as well, since that is an expected courtsey.)Without the second action, the first is sterile and almost evading responsibility.

Yes, none of the options are easy, and they involve either inviting backlash or carrying guilt. One must take a decision based on one's fortitude and appetite for criticism. I personally cannot recommend taking the softer way out, so my recommendation would be either (3) or (4) depending on how strongly one feels about the issue.

EDIT:

Following a suggestion in the comments, it may be worth differentiating the background of (3) and (4).

In (3), where one conveys one's apprehensions through a private communication to the editor, the moral burden is placed squarely on the editor/journal. The reviewer has done a purely academic duty and is satisfied with communicating apprehensions privately. There could be some good reasons to adopt this stance- trusting the ethical/moral core of the editor/journal, believing that this grey area should be dealt with within professional confines and not publicly.

In (4), one does not absolve oneself on any moral responsibility, and chooses to tackle ethical concerns personally. This means accepting responsibility not only for backlash, but also for the successful dissemination of the message. It is inherently the more difficult, but more responsible and possibly satisfying approach.

If it really troubles you, withdraw from reviewing the paper and tell the editor about the ethical concerns.

If you reject the paper for anything other than academic reasons, you're implicitly imposing your ethics on the authors. Ethical statements that make perfect sense to you might not make sense at all to someone else; the very fact that the authors did the research indicates they don't find autonomous armed drones ethically abhorrent. You could claim that your ethics are "better" than theirs, but that is not certain, and a reasonable person could take the reverse position. If you want nothing to do with the paper for ethical reasons, decline to review it, and let someone who is less troubled by the ethics review it.

That said, you can point out the issue to the journal and make sure they are OK with the ethics. It's possible the journal will agree with you that autonomous armed drones are unethical and will reject the paper without review. If they publish anyway, you can distance yourself from the journal by not reviewing for them and/or publishing with them in the future.