Daniel Berend

Given a binary sequence, one may inquire whether it is produced by a true random source. There are several tests designed to answer this question, such as the statistical test suite of the National Institute of Standard and Technology (NIST) and the Diehard tests. The problem is that, given deterministic tests of randomization, an adversary may know/learn, the adversary may tailor a non-random (deterministic) sequence, guided by the deterministic tests, that passes the tests. We suggest to use a true random source for randomness tests and thus make the tests significantly harder to being misled. We design tests that use true random sources and demonstrate their ability to detect non-random sequences that NIST classifies as random.