Abstract: This study aims to discuss the risks and opportunities involved in building predictive models based on artificial intelligence. Countermeasures are also proposed to minimize the risks involved in their adoption where reliability is a critical factor for user safety such as autonomous driving. For this, it is explored a real development of a predictive mathematical model, using industrial data in the steel industry. This development aimed to construct an empirical mathematical model to predict the mechanical properties (Yield Strength, YS) of hot rolled steel structural beams. Such model was based on rolling process variables and the chemical composition of steel. As a result of this research it was observed that the obtained data agreed with the expected metallurgical theory. The errors obtained between the estimated and the real values were greater for process conditions with lack of enough data. These results are associated with the risk of using artificial intelligence technology in critical applications and actions aiming at its improvement are proposed.