A Transformer-based Text Classifier Using Sticky Policies to Reduce Confidentiality Obstacles in Enterprise GenAI Systems
Open Access DepositedPrivacyBridge
Downloadable Content
Generative Artificial Intelligence (GenAI) is transforming the industry in different sectors, including healthcare, finance, and technology. Organizations are leveraging GenAI’s capabilities to streamline their operations. In stark contrast, GenAI presents significant confidentiality risks to businesses. Research indicate that sensitive proprietary and personal data from multiple organizations are getting exposed through user prompts. Exposure of sensitive information through prompts is becoming a contributor to the broader issue of information loss, which costs U.S. companies up to $600 billion every year. As a result, twenty five percent of early adopters of GenAI have opted to limit its usage in their environments. This praxis addresses the urgent confidentiality challenge enterprises face when adopting GenAI. It introduces a new text classification model dubbed PrivacyBridge, which employs a fine-tuned Text-to-Text Transfer (T5) transformer model along with sticky policies. This research not only contributes to the development of an improved model for detecting and classifying sensitive enterprise information within user prompts but also demonstrates the effectiveness of integrating sticky policies into transformer models. The proposed model significantly outperforms existing sensitivity detection models, achieving an accuracy of up to 95.45% and a precision of up to 90.91%.
- All rights reserved
Notice to Authors
If you are the author of this work and you have any questions about the information on this page, please use the Contact form to get in touch with us.