Kant is usually interpreted to have two formulations. First, “Act so that the maxim of your action, or the generalization describing it, can be willed by you to be a general rule, to be followed by all rational agents.” In other words, only do those things that you as a benevolent, rational being can will that everyone do.

The test of a categorical imperative is what happens if everyone follows it. Something that has good consequences in a particular case might not have good consequences in all cases. For example, if the maxim is “Obey traffic rules,” and you come to a red light with no other cars in attendance, you may not drive through it, even though the consequences in this particular case would be benign. Or, to use an example of Kant’s, if the maxim is not to lie, and a madman is looking for a friend of yours whose whereabouts you know, you may not lie in this case, because overall you can’t benevolently will that everyone be permitted to lie whenever the consequences are good for them. To take another example of Kant’s, you may not take your own life, no matter how miserable you are, because you categorically can’t will suicide as a good action.