There are quite a few things to unpack here, so let’s go over them one by one:
You are responsible for delivering an accurate report; you are not responsible for indirect consequences of an accurate report, such as career impact.
If somebody does not get a position because you gave a correct negative assessment, this is upon them or their co-authors.
Also remember that somebody else will get the respective position instead and – on average – they are more qualified:
There is a lot of noise in career decisions and by providing a fair review, you reduce this.
Do not only think of the authors, but also think of the qualified researcher who does not get a position because their opponent managed to get some piece of junk published thanks to overly lenient reviews.
If you want to be nice to upcoming researchers without compromising the quality of research, do your peer reviews quickly (i.e., prioritise them with respect to your other duties, not rush the actual reviewing). This way, they have more time to fix the paper’s problem and submit it elsewhere before whatever application deadline they are facing and they hopefully learn something in the process:
Coping with and learning from rejections and criticism in general is an important skill for researchers.
If a student’s supervisor did not properly guide them through their research or writing process, the student has to learn this another way.
You are not responsible for taking that part of the training, but a detailed review can be that way.
Finally, it may also help you to assume that all the authors are professors close to retirement with no career stakes whatsoever.
The ideal of peer review is that it should ignore the authors’ identity (which is why double-blind peer review is a thing).
From the reviewer’s and editor’s point of inevitably imperfect knowledge, every paper has the potential to be salvaged: The authors may fix their research or explanations or present better arguments for the relevance of their work. The probability that a paper can be made suitable for a journal is never zero. Going by this, there never should be a decision for final rejection. But this is not what a rejection practically means.
Instead, if this probability (of the paper eventually being made suitable for the journal) turns out to be below some implicit threshold, it is more feasible to reject the paper and not invite further submissions, which would waste further journal resources.
All you do is to help estimate this probability and whether it is below the threshold.
If your estimate is a little bit off, that’s not dramatic, as there are several mechanism to compensate for this, namely other reviewers, the editors, and the option for rebuttal, if the authors think that they solved the issues with the manuscript against all odds.
What’s more important is that you elaborate what problems you see with the paper and what needs to be done to solve this (Ian Sudbery’s answer gives a good example for this).
Even if your criticism should be based on some misunderstanding, this allows the authors to clarify.
Nothing is more frustrating if the main argument for rejecting your paper is something like:
I don’t understand Section 2. [No further elaboration]
as you have to resort to guessing on how to solve this.
Which brings us to your next concern:
Too Detailed Reviews
I do not think details are harmful in a review per se.
The main reason to stop delving into details is to avoid wasting your time.
Of course, you should make clear what the main issues are that lead to the rejection.
A detailed review can be very helpful to the authors to solve these main issues, be it for a resubmission or submission to another journal.
Moreover it provides evidence to the editor that you seriously engaged with the manuscript thus giving your review deservedly more weight.
Keep in mind that almost every manuscript gets published eventually, so your effort is rarely wasted (unless you do such things as noting typos in paragraphs that need to be completely rewritten anyway).
You may note that you recommend to reject papers more often than your own papers are rejected or you even hear of papers being rejected.
There is a good chance that this says more about the quality of your and your acquaintances’ submissions than about your recommendations as a peer reviewer.
Remember Sturgeon’s law:
“90% of everything is crap.”
It may be worth checking out the actual rejection rate of the journal if you can find it.
But even if your rejection rate is higher than the general rejection rates of the journals you are reviewing for, this is not necessarily a bad thing.
First of all, there is the matter of chance:
You have to review quite some papers to make any substantiated statements about your reviews deviating from the mean.
Second, the more thorough a review is, the more likely it is to find a flaw leading to a recommendation of rejection.
Here, the problem is not necessarily you but other reviewers not being thorough enough to spot problems.
All reviews I have received, written, and co-written were almost exclusively in a very neutral tone, i.e., neither overly euphemistic nor aggressive.
There are justified exceptions, e.g., I once wrote in a report that I was disappointed by a revision on account of not implementing straightforward suggestions with which the authors appeared to agree.
If you stick to a neutral description of issues and suggestions, I see little that could go wrong.
The one exception where wording matters is the actual recommendation – if this happens in writing and not via a score system.
Here, you do not want to be misunderstood by the editor.
The aforementioned list of things that need to be happen before acceptance is a good way to avoid any miscommunication here:
For example, if the editor can easily see that this is probably an insurmountable challenge, they do not need your direct assessment.
It may also help you to write a short note exclusively for the editor explaining how likely you think that the authors manage to fix their manuscript.