In this blog post we present a Bayesian statistical model to detect cryptographic timing attacks. This model is one of the results of a customer hardware assessment performed by the SCHUTZWERK GmbH. The assessment was performed in a gray box context, i.e., we were able to interact with the encryption hardware, but were not given any internal implementation details.
In software dealing with security, randomness is often necessary to generate keys or tokens for resetting passwords or identifying sessions. There, randomness is required to be unpredictable for an attacker. However, sometimes developers do not use cryptographically secure pseudo random number generators (CSPRNG) in this scenario. Instead they utilize faster pseudorandom number generators (PRNG). Consequently, the question arises how hard it is to attack a common (not cryptographically secure) random number generator.