In a world where most of our choices and decisions are data-driven, information systems become an asset. Any damage to this asset can leave a huge impact with unforeseen consequences. This is why cyber security is finding interest among the information technology and busi ness communities. Day by day, people are becoming more aware that without protecting the flow of information used daily, the progress we have can fade away. To assess the security of their information systems, organizations use ”Penetration tests” to detect the vulner abilities in their systems and to which extent they can be exploited. Also called Ethical Hacking, it aims to simulate the activities of a real hacker and evaluate the chain of events in case of an attack. Further more, technology is at a fast pace of advancement, attackers now have the benefit of using Machine Learning (ML) to conduct more danger ous attacks. To fight back, penetration testers also need to leverage the power of machine learning to conduct smarter penetration tests with more automated phases eliminating factors that can be faced by humans such as tiredness and human errors. In this thesis, we evalu ate the importance of ML in penetration testing and propose a model to detect malicious URLs based on Deep Learning variants (CNN, LSTM, bidirectional LSTM), then compare these variants in terms of accuracy and precision, where CNN with bidirectional LSTM pro vides the best results. This solution in this thesis proves that Machine Learning is a big support to penetration tests.