Phishing-emails are the most powerful cybersecurity attacks because they are inexpensive and linguistically advanced. This paper describes a new method for detecting phishing emails based on AI and machine learning techniques. A unified dataset of 481 unique messages was compiled from numerous public sources of phishing messages after a series of deduplication operations. The natural language content is preprocessed in a standard manner. Relevant data was extracted in the following manner: The feature extraction technique combines a set of ngrams based on the TF-IDF and eleven manually constructed linguistic and url-based features, yielding a 261-dimensional feature set. Seven machine learning classifiers and the ensemble of these classifiers were trained and tested. Specifically, we used the ‘Logistic Regression,’ ‘Random Forest,’ ‘Extra Tree,’ ‘Gradient Boosting, ‘Catboost,’ ‘Supporting Machine,’ and ‘Naive Bayes.’ A series of experiments was carried out. From the experimental results, the proposed system is more effective than other classifiers and has a testing accuracy as high as 90.72%. Recall is a very important issue in security and is handled in the proposed system