Suppose that I wrongly steal some wallet from some woman dressed in white who is eating strawberries while reading the last page of Spinoza's Ethics. My maxim is [unrealistically, but let's suppose] to act in precisely this way, whenever I can. I could rationally will it to be true that this maxim is universal, because it would be most unlikely that anyone else would ever be able to act in precisely this way, so this maxim's being universal would be most unlikely to make any difference. Since I could rationally will this maxim to be universal, Kant's formulas mistakenly permit my act.
Parfit then suggests that Kantians can avoid this objection if they reformulate their principles by replacing any mention of the agent's maxim with the more objective standard of relevantly similar circumstances. If we can't rationally will that others act similarly in morally similar circumstances, then that's enough to render the act impermissible, on this version of the view. Parfit explains:
When we apply these formulas to someone's act, we must describe this person's act in the morally relevant way. Suppose that, being a whimsical kleptomaniac, I really am acting on the maxim of stealing from white-dress-wearing strawberry-eating women, whenever I can. This maxim does not provide the morally relevant description of my act. It is irrelevant that I am stealing from someone who is a woman, and who is wearing white and eating strawberries. The relevant facts may be that I am stealing from someone who is no richer than me, merely for my own amusement. [And I plausibly could not rationally will that this class of actions be universalized.]
There's something dialectically fishy about this. Parfit has previously emphasized that we can only appeal to non-deontic reasons if we are to avoid trivializing the Kantian formulas. It would be viciously circular to say, e.g., that lying is wrong because you can't rationally will that everyone lie, and that the reason why you could not rationally will that everyone lie is because lying is wrong. The latter claim requires that we have some antecedent explanation of why lying is wrong, before we apply the categorical imperative. So the categorical imperative would not then be the fundamental principle of morality. Instead, we would be fundamentally relying on the antecedent principle (whatever it was).
But isn't the same thing true here? Parfit is relying on some antecedent specification of "moral relevance". Since this is an input to the Kantian formulas, the latter cannot be what fundamentally ground this moral status. But then we need some more fundamental moral theory to explain which features of acts are the morally relevant ones, and hence fix which of the many possible descriptions of an act is the morally relevant description.