The safety of Automated Vehicles (AV) as Cyber-Physical Systems (CPS) depends\non the safety of their consisting modules (software and hardware) and their\nrigorous integration. Deep Learning is one of the dominant techniques used for\nperception, prediction, and decision making in AVs. The accuracy of predictions\nand decision-making is highly dependant on the tests used for training their\nunderlying deep-learning. In this work, we propose a method for screening and\nclassifying simulation-based driving test data to be used for training and\ntesting controllers. Our method is based on monitoring and falsification\ntechniques, which lead to a systematic automated procedure for generating and\nselecting qualified test data. We used Responsibility Sensitive Safety (RSS)\nrules as our qualifier specifications to filter out the random tests that do\nnot satisfy the RSS assumptions. Therefore, the remaining tests cover driving\nscenarios that the controlled vehicle does not respond safely to its\nenvironment. Our framework is distributed with the publicly available S-TALIRO\nand Sim-ATAV tools.\n