The high rate of artificial intelligence creation of text has posed significant issues on the maintenance of content authenticity, academic integrity, and control of misinformation. This paper presents a deep learning style model of classifying human written and AI generated text using a Bidirectional Long Short Term Memory BiLSTM. The model was trained and tested on a large scale dataset of 44,898 samples which includes 23,482 AI generated and 21,416 human written texts. The dataset contains both structured metadata (title, subject and date) and unstructured textual content, which can be analyzed in a comprehensive way through Natural Language Processing based analysis. Common preprocessing methods, like tokenization, vocabulary building, and sequence padding, were implemented to convert textual data into appropriate numerical data. In order to have a consistent and reliable performance evaluation and to minimize the chances of data leakage, a stratified data splitting strategy has been adopted to have similarity in class distribution between training and testing sets. The experimental findings proves that the suggested BiLSTM model recorded a classification accuracy of 99.93 percent with minimal validation loss, which means that it possesses a high ability to identify bilateral contingency and semantic patterns of text sequences. Bidirectional architecture increases the model capability of learning complex linguistic structures, and the reliability of classifications. The given solution can be a scalable and efficient method of automated detection of AI generated content, and it can be used in academic integrity checking, misinformation, and digital content verification. The research findings indicate the usefulness of deep learning founded sequence modeling to tackle emerging issues in AI generated text identification.