This paper uses neural networks to classify texts related to harassment and discrimination in various contexts, such as workplace harassment, school bullying, and sexism. In this research, it is proposed to address harassment in different environments through an approach based on Long Short-Term Memory (LSTM) models. The main objective is to develop a model capable of identifying patterns of harassment in texts, improving detection accuracy, and reducing biases present in traditional methods. By using LSTM networks, the model can process text sequences and associate multiple relevant labels, improving the detection of subtle and complex forms of harassment. This work contributes to efforts to combat harassment by providing a solution that can help better understand and address these issues in texts derived from real-world experiences. The results obtained show a precision of 0.8212, a recall of 0.6883, an F \(_1\) of 0.7489, and an accuracy of 0.7782, confirming strong performance in classifying texts related to harassment and discrimination.