The generation of random numbers is a keystone function in any cryptographic protocol. Indeed, in a security context, the random numbers generation shall withstand assaults from adversaries. It is thus paramount to validate both its functionality and its robustness in front of attacks, including fault injection attacks. The verification implies tests, which shall thus be carried out in nominal but also in perturbed operational environments. In this paper, we review standard tests already existing and still under development. As a first contribution, we suggest a new kind of metrics to assess the quality of the random sequences of bits. As a second contribution, we analyse fault injections in true random number generators and explore whether such faulted behavior can be self-induced within the circuit itself. This analysis reveals a plausible interpretation of the behavior of circuits based on the analysis of long term noise, e.g., TRNGs based on ring oscillators.