Adversarial machine learning research has revealed that algorithms can be misled. The research has shown that this may cause misinterpretations of traffic signs by autonomous cars. Yet, legal literature has not yet dealt with this issue. This research will incorporate the findings of adversarial machine learning and question the traditional concepts of liability. These traditional concepts may prevent innovation by putting too much liability on the developer. Therefore, the research will argue for a more flexible regulatory regime and eventually for considering giving sui generis persoonhood to algorithms. We presume that regulating alhoritm is complicated due to the uncertainty on the risks. The research will engage desk research, interviews, and experiments to achieve these results.