So, my thought process here was: given a fairly robust selection of certificate authorities, a given user could have a number of different client certs for use in different trust scenarios. Contrast the following:
- A user bearing a client cert with the name "Jonathan Grant", signed by a U.S. government agency which is known to verify that its signees' certs are a citizen of the United States.
- A user bearing a client cert with the name "Michael Black", signed by Alice, who is known to only sign certs after verifying that the real-world identity of the signee matches the name on the cert.
- A user bearing a client cert with the pseudonym "c00ln4m3", signed by Bob, who is known to only sign a single cert for any given real-world person. (To do so, he verifies the person's real-world identity but does not reveal which cert corresponds to which person.)
- A user bearing a client cert with the pseudonym "hunter217", signed by Charlie, who is known to sign certs without verifying the real-world identity of his signees at all, but who is also known to revoke his signature on certificates if a service provider complains about the user bearing that cert.
- A user bearing a client cert with the pseudonym "cypr3ss", signed by David, who is known to charge $1000/year for a cert bearing his signature but performs no other identity verification.
The point of listing out these different scenarios is that the underlying mechanism (client certs) is the same, but the end-user and the service provider don't actually have to trust each other: they only have to agree on a CA with mutually acceptable policies.