it is a misuse of the word 'ought' to say 'You ought, but I can conceive of another situation, identical in all its properties to this one, except that the corresponding person ought not'. (MT p.10)
This places certain constraints on sincere moral judgment. You can't think it's okay to hurt another, unless you also endorse being hurt yourself when in their situation. But what if you will endorse this? Consider the fanatical Nazi, so committed to the cause that he's willing to prescribe his own extermination were he to turn out to be a Jew himself. Is Hare then committed to saying that the fanatic makes no error when he says that all Jews should be exterminated?
I think his view does entail this, though Hare himself thought otherwise. Universalizabilty requires that we can endorse the outcome even after imagining ourselves in others' positions, with their desires, preferences, and moral ideals. Crucially, Hare thought that in doing this, we would inevitably come to share their preferences about the situation they are in. So, in thinking about what we can endorse after considering all positions equally, we must ultimately prescribe preference utilitarianism: maximizing the net weight of preference satisfaction across all involved.
However, Hare's 'crucial step' seems simply mistaken. It conflates counterfactual desires with desires about counterfactuals. Note the difference:
(1) In situation S, I would desire that P
(2) I do desire that, were I in situation S, P.
So merely imagining a situation where I have some desire, doesn't entail my (now) endorsing it. The fanatical Nazi can say, "Wow, yeah I can see I'd feel really unjustly abused if I were a Jew. I'd certainly oppose the Final Solution. How misguided I would then be! Such unpatriotic preferences should not be given normative weight."
Peter Singer ('Reasoning towards Utilitarianism') notes that Hare's non-cognitivism may be used to undermine this last claim, for it means that no personal preferences/ideals are intrinsically superior to any others. So the fanatic cannot ground his stubborn preference on claims of objective truth or warrant. Fair enough. But did he ever need to? For non-cognitivists, preferences don't need to be 'grounded' or justified at all. So why not a higher-order preference to the effect that such-and-such other preferences -- even if they come to be one's own -- should be thwarted? It may be an uncomfortable preference to sincerely hold, but if the fanatic is willing to really bite this bullet, then I don't see how Hare has the resources to criticize him.
It's important to note that the fanatic really is making a universal prescription here. He's not just claiming that anti-Nazi preferences should be thwarted given that only other people have them. He's also willing to endorse this prescription for all possible situations that are relevantly similar, including that possible situation where he himself is anti-Nazi. This willingness to universalize is precisely what makes him a fanatic rather than an amoralist. Granted, when he fully and vividly imagines being Jewish himself, part of what he imagines is opposing the prescription of Nazism. But he can imagine opposing a prescription without thereby actually opposing its application to the imagined scenario, as we saw in distinguishing (1) and (2) above.
This is exactly like the point that I can imagine a scenario where I [perhaps falsely] believe that P, but it doesn't follow that I (actually) believe, of that scenario, that P is true in it. We must take care to separate the content of the scenario being imagined (including the attitudes I am stipulated to have within or according to the fiction) from the attitudes I have towards the imagined scenario. Once we're clear about this, I don't see how the non-cognitivist can accuse the fanatic of error in his (non-utilitarian) universal prescriptions.
Update: It's been pointed out to me that Hare was actually careful to distinguish (1) and (2). He just thought that a full appreciation of the first would necessarily bring about the second -- otherwise one must not have been imagining the counterfactual preferences vividly enough, or some such. That doesn't strike me as a plausible claim, though.