Abstract Background Prevention of violent behaviors (VB) in the early phase of psychosis (EPP) is a real challenge. Impulsivity was shown to be strongly related to VB, and different evolutions of impulsivity were noticed along treatments. One possible variable involved in the relationship between VB and the evolution of impulsivity is cannabis use (CU). The high prevalence of CU in EPP and its relationship with VB led us to investigate: 1/the impact of CU and 2/the impact of early CU on the evolution of impulsivity levels during a 3-year program, in violent and non-violent EPP patients. Methods 178 non-violent and 62 violent patients (VPs) were followed-up over a 3 year period. Age of onset of CU was assessed at program entry and impulsivity was assessed seven times during the program. The evolution of impulsivity level during the program, as a function of the violent and non-violent groups of patients and CU precocity were analyzed with linear mixed-effects models. Results Over the treatment period, impulsivity level did not evolve as a function of the interaction between group and CU (coef. = 0.02, p = 0.425). However, when including precocity of CU, impulsivity was shown to increase significantly only in VPs who start consuming before 15 years of age (coef. = 0.06, p = 0.008). Conclusion The precocity of CU in VPs seems to be a key variable of the negative evolution of impulsivity during follow-up and should be closely monitored in EPP patients entering care since they have a higher risk of showing VB.