Fake news not only misleads the readers but also creates a devastating impact in society so it is a major concern in today’s digital world where proper news is very important and crucial. The increasing fake news globally tends to encroach human right and public safety. A person needs to have vast knowledge about the current issue to discover whether a news is real or fake. So, we have implemented a system to detect the fake news using Natural Language Processing for dataset preprocessing, Bi-directional LSTM using Word2Vec vectorization to predict the news. Other models like Bi-LSTM and LSTM with one hot encoding and Word2Vec are also used for comparing accuracies and performance. With number of tests on both Bi-LSTM and LSTM models using one hot encoding and Word2Vec on each model as vectorization, the best model was found to beBi-LSTM Model where Word2Vec was used with testing accuracy of 99 percent. Here we only worked on one specific domain of news viz. politics mainly from the USA.