The Cosmic Microwave Background (CMB) maps measured by the PLANCK experiment will allow us to test the physics of the primordial Universe with unprecedented accuracy. In particular, the detection of a deviation from Gaussianity in the CMB would rule out the simplest models of cosmic inflation. Unfortunately, several late-time mechanisms lead to non-Gaussian features in the CMB, thus inducing a bias on the estimation of the primordial signal. In our work, we focus on non-linear dynamics, whose contamination to the primordial non-Gaussianity has not been quantified unambigously in the literature. As non-linearities arise naturally in General Relativity, it is crucial that they are well-understood and hence removed from the data. In order to do so, we have written a second-order Boltzmann code to compute the CMB bispectrum, which involves the formidable task of solving the Boltzmann-Einstein system of equations at second order in perturbation theory. In this talk, I shall summarize the previous literature on the topic, highlight the differences with the first-order approach, describe the status of the project and present some preliminary results.