MUMBAI, India, May 1 -- Intellectual Property India has published a patent application (202641051773 A) filed by MLR Institute Of Technology, Hyderabad, Telangana, on April 23, for 'hybrid bert-bilstm-attention architecture for high-precision detection of ai-generated text.'

Inventor(s) include Mr. V. Sai Krishna; Ms. Balla Keerthana; Ms. Bommala Divya; and Ms. Desireddy Mahitha Sai.

The application for the patent was published on May 1, under issue no. 18/2026.

According to the abstract released by the Intellectual Property India: "This invention presents a hybrid neural architecture designed for high-precision detection and classification of AI-generated text in diverse digital environments. The proposed system integrates a fine-tuned BERT-base model to extract deep bidirectional semantic context, enabling comprehensive understanding of token-level and sentence-level relationships within textual data. To further enhance structural and sequential comprehension, a Bidirectional Long Short-Term Memory (Bi-LSTM) network is incorporated to capture forward and backward temporal dependencies, thereby modeling contextual continuity and linguistic flow. A global attention mechanism is applied to refine the extracted features by dynamically assigning weighted importance to the most informative semantic and structural components, ensuring suppression of irrelevant noise and amplification of discriminative signals. The architecture further employs a calibrated sigmoid-based decision layer with an optimized decision boundary threshold of 0.80, significantly improving classification confidence and achieving an overall accuracy of 94.19%, surpassing conventional standalone transformer models. The system is specifically engineered to address critical real-world challenges including misinformation detection, academic plagiarism prevention, authorship verification, and identification of content generated by advanced large language models such as Claude and GPT. Through multi-layer contextual fusion, adaptive threshold optimization, and robust feature learning, the invention provides a scalable, reliable, and high-performance framework for AI-text authenticity verification in modern information ecosystems."

Disclaimer: Curated by HT Syndication.