The title may read like a contradiction, but I would like to know whether a cryptographic solution exists to the following problem:
Suppose Alice wants to write a document using a pseudonym, so that, in general, people cannot connect her with her work. However, Alice may want to convince Bob (with at least a high level of confidence) in certain situations that she indeed is the author. At the same time, Alice does not want Bob to be in possession of a proof linking the author to the work. That is, Alice does not want Bob to be able to convince Charlie that Alice is the author of the document.
This may be useful in a setting where, say, a programmer might wish to publish code, or information pertaining to security research, without publicly associating himself with the information. Nevertheless, when hunting for a new job, the author may want to convince prospective employers of his abilities, but not give them proof that could de-anonymize him to the public at large.
I am not a cryptographer, but here are my thoughts so far:
Alice chooses a pseudonym, along with an RSA keypair, and sets up a public repository on e.g. github, which acts as a trusted third party, proving that certain data had existed at certain points in time, and recording all public manipulations thereof. Alice signs all her commits with her private key, and at some point interviews with a new employer.
Alice might offer to log into her pseudonymous github account and make publicly visible changes, but the interviewer might argue that she shares her login info with someone else, who might use the account in other interviews.
So I guess that Alice would have to upload data early in her publication chain that is advertised as the only and definitive proof of identity (such that there are not multiple hidden proofs for various real names). She may, for example, upload a cryptographic hash of a string that contains her name, followed by a secret random string known only to her. However, if she gave that string to an interviewer, the interviewer might save that string and show it to others, who then would also be convinced of Alice's authorship. Alice could offer to enter the secret string into devices only she controls, but then the interviewer might object, as he has no proof that the program really does compute the intended cryptographic hash.
Is there a way to achieve what I am after? The fewer assumptions necessary (e.g. trusted third parties), the better!