The approach presented here involves the detection of AI-generated text across languages by using a BERT-based multilingual cased model. This model is tested on independent samples retrieved from English and Hindi. Testing the model on these samples retrieved its key features such as lexical diversity, syntactic complexity, and semantic coherence. Fine-tuning of these features is done, allowing the model to make a difference between human-written text and machine-generated text in both languages. Further to this, the work assesses the translatability of the approach into low-resource languages and analyses its performance on various genres of text. This approach significantly changes the game regarding how to best preserve the integrity of text content in contexts involving moderation, academic integrity, or journalism. The approach, if adopted, gives the community at large a sound resource and tool for conducting challenging analysis and research as natural language generation technologies are furthered by advanced development in them.