The swift progression of technology has exacerbated the dissemination of disinformation, especially on digital and social media platforms. This study examines machine learning methodologies for identifying deception generated by both artificial intelligence and humans. We assess logistic regression (LR), support vector machines (SVM), decision tree classifiers, and an ensemble model that integrates their predictions. The decision tree model attains an accuracy of 97.5% in differentiating real news from AI-generated content, whilst the ensemble model enhances performance to 98.2%. The ensemble model attains an accuracy of 92.5% in distinguishing authentic news from human-generated false news, surpassing the performance of individual classifiers. The ensemble technique demonstrates an accuracy of 93.2% in addressing all forms of deception. To augment robustness, we expand our dataset to incorporate content from GPT-2, GPT-3, and ChatGPT, illustrating the increasing complexity of synthetic text. We utilize a transformer-based model (BERT) that outperforms conventional models, attaining an F1-score of 0.91. These findings highlight the efficacy of utilizing both ensemble and transformer-based models for enhanced generalizability and semantic comprehension. We emphasize the practical importance of our method in facilitating real-time fact-checking and automatic content filtration. The modular architecture of our system enables adaption to various disinformation formats and linguistic differences, making it a scalable solution for the ongoing challenge of detecting fake news.