It is widely believed that there should be no solutions available, even privately, since this somehow ruins the game. This presumes that there should be "exercises" of the traditional sort in advanced mathematics courses, which is already partly dubious, since (as is often visible in commercially successful texts) it leads to make-work exercises often of questionable interest. I'd agree that there do exist significant, meaningful questions that may not fit into a small book... but would argue that then good write-ups of their solutions/resolutions should be available somewhere as models. Otherwise, all the students ever see is their peers' solutions... which in principle could be fine, but, observably, in practice, often overlook (through misunderstanding) ideas (from the text or otherwise) that make the resolution far more graceful and persuasive. That is, without good solutions, the only models anyone ever sees are "iffy".
(E.g., my abstract algebra text originally aimed to work a large fraction of the traditional significant questions as "examples", exactly to overcome the inertia of traditional-not-so-good alleged solutions of them, and have no "exercises" whatsoever. However, the publisher, who'd already made surprising concessions about intellectual property stuff, really-really wanted "exercises". So I made some near-clones of the worked exercises... And I've received several comments that I'm an anarchist for making those good solutions public!)
So, indeed, I think it's a bad idea to try somehow to suppress "good solutions". People will still grasp at bad solutions, and will be learning deficient versions of things to the extent they learn anything.
By the way, it is certainly not the case that the standard graduate mathematics texts provide means to resolve all their exercises. Often there is a considerable disjunction. Typically, the disjunction is that the theorems in the chapters do not at all suggest any quasi-algorithmic devices for doing computations in any particular case. E.g., abstract Galois theory usually disregards Lagrange resolvents, so does not hint at how to solve equations even when they can be proven solvable by radicals...
Nor is it the case that beginning math grad students are adepts at writing... so there is considerable feedback among them of marginal write-up style, marginal technical viewpoint, too much attention to secondary and tertiary details (often strictly demanded by in-my-opinion misguided texts or instructors), and needlessly distorted ambient language. Good writing models would help people "get over" this.