We report the observation of the tttt production in LHC collisions, performed on the data collected by CMS during Run2 (2016-2018). The production of tttt decays is one of the most rare processes at the LHC and among the most complex and rich signatures to study due to the large jet and lepton multiplicity in its final state. Leveraging the improvement in event reconstruction and improved b-tag and lepton identification algorithms, based on machine learning, the expected sensitivity to tttt production in events with at multiple leptons is substantially improved with respect to our previous results.