These are hierarchical in the way that genus, species and individual are hierarchical.
- Subject - In a security context, a subject is any entity that requests access to an object. These are generic terms used to denote the thing requesting access and the thing the request is made against. When you log onto an application you are the subject and the application is the object. When someone knocks on your door the visitor is the subject requesting access and your home is the object access is requested of.
- Principal - A subset of subject that is represented by an account, role or other unique identifier. When we get to the level of implementation details, principals are the unique keys we use in access control lists. They may represent human users, automation, applications, connections, etc.
- User - A subset of principal usually referring to a human operator. The distinction is blurring over time because the words "user" or "user ID" are commonly interchanged with "account". However, when you need to make the distinction between the broad class of things that are principals and the subset of these that are interactive operators driving transactions in a non-deterministic fashion, "user" is the right word.
Subject/Object inherits from the same terms as used in grammar. In a sentence the subject is the actor and the object is the thing acted on. In this sense the use has been around since before computers were invented. In a security context, a subject is anything that can make a request. As noted above, this need not be limited to IT security and so is a very broad classification. The interesting thing is that subject implies object. Without an object, there is no subject.
Principals are what subjects resolve to. When you present your credit card you are the subject and the account number is the principal. In other contexts your user ID or state-issued identification is your principal. But principals can be associated with many types of subject that are not people. When applications make requests for system-level functions the principal may the signer of a signed executable code module but even in that case the user driving the request is still the subject.
User is more specific than subject or principal in that it usually refers to an interactive operator. That is why we have a graphical User Interface and not a Graphical Principal Interface. A user is an instance of subject that resolves to a principal. A single user may resolve to any number of principals but any principal is expected to resolve to a single user (assuming people observe the requirement not to share IDs). In the example above, the signer of an executable code module is definitely not the user, but it is a valid principal. The interactive operator trying to get the module loaded is the user.
As noted in the comments, even the authoritative sources do not agree on these terms. I searched NIST, SANS, IEEE, MITRE and several "quasi-authoritative" sources such as security exam guides while preparing this response. No single source that I found which was at least quasi-authoritative covered all three terms and all differed significantly in their usage. This is my take on how the terms should be used but from a practical standpoint, when you are poring over a manual in the middle of the night, the definitions tend to be whatever the vendor or writer say they are. Hopefully though responses here will provide enough insight to navigate the waters and parse any security document using these terms.