If murder were merely an agent-neutrally bad type of occurrence and nothing more, then the badness of one murder would be outweighed by the badness of two or three others, and one could be justified in murdering one innocent person to prevent three others from being murdered. But if there is a right not to be murdered, it does not give way when murdering one innocent person is the only means of preventing the murder of two or three others. A right is an agent-relative, not an agent-neutral, value: Rights tell us in the first instance what not to do to other people, rather than what to prevent from happening to them.

It is compatible with this conception of rights that they are not absolute, and that there may be some threshold, defined in consequential, agent-neutral terms, at which they give way. For example, even if there is a general right not to be tortured or murdered, perhaps there are evils great enough so that one would be justified in murdering or torturing an innocent person to prevent them. But this would not change the basic character of the right, since the threshold will be high enough so that the impermissibility of torture or murder to prevent evils below it cannot be explained in terms of the agent-neutral badness of torture or murder alone. Even if it is permissible to torture one person to save a thousand others from being tortured, this leaves unexplained why one may not torture one to save two.

(Thomas Nagel, “Personal Rights and Public Space,” Philosophy & Public Affairs 24 [spring 1995]: 83-107, at 88-9 [italics in original])