The interest in autonomous navigation for surface ships has arisen over the years for both naval vessels and commercial ships. In literature, several pieces of research addressed the challenges of developing new guidance laws, stable and robust control algorithms, methodologies to increase situational awareness, and collision avoidance algorithms. Current papers base their outcome on blind trust in the information coming from navigation sensors. However, sensors can be subjected to malfunctioning, or even worse, cyber-attackers may hijack their data. This last scenario will be more and more common in the near future and represents a dangerous threat that future generation ships must face. In this work, a grey-box approach has been used to predict the outcomes of a state-of-the-art collision avoidance algorithm. Based on that, the attack consists of injecting fictitious targets as the input of the collision avoidance to force the ship to follow a predetermined and malicious track. A set of dedicated simulations using a ship simulator is carried out, and the effects of such a cyber-attack on the automatic collision avoidance system are shown. Based on the obtained results, at the end of the paper, effects on the evasive route generations are analysed and deeply discussed.