A secure human-computer authentication scheme

We introduce a protocol for authentication between a human and a computer, where the human is able to use no special hardware other than a dumb terminal. Authentication is based on a shared secret which can be reused polynomially often with no danger of exposure, assuming the conjectured uniform hardness of learning parity functions in the presence of noise. Under this conjecture, the protocol is secure against a polynomially-bounded passive adversary and also some forms of active adversary, although it is not secure against arbitrary active adversaries. University Libraries Carnegie Mellon University Pittsburgh, PA 152 i 3-3890