Fair Mutual Authentication

We consider a fair authentication process where at each moment of the protocol execution each participant has almost the same certainty about the identity of the other participant. We combine this property with authentication with errors: each authentication bit may be replaced to the wrong value. Thereby, an observer attempting to derive the secret key(s) used for authentication in a cryptanalytic way has substantially harder job due to an unknown error pattern (learning secrets with errors). We show that the presented protocol satisfies subtle requirements of the GDPR Regulation of data minimization in case of failure.