Session 5-F
Preventing accidental data leaks using LLMs
In the era of growing unstructured data and advanced Large Language Models (LLMs), the risk of sensitive data exposure has become a paramount concern for enterprises. With an average of 17.5 petabytes of unstructured data, including 55% classified as dark data, organizations face the challenge of safeguarding data in AI interactions. In this talk, we will present a solution architecture that integrates AI-driven data classification, robust access controls, and compliance mechanisms. We will describe how this approach enhances data security, ensures AI compliance, and streamlines sensitive data management while boosting operational efficiency and risk mitigation. We will also discuss why it is pivotal for organizations seeking to leverage AI capabilities responsibly in the LLM era to adopt this framework. Implementing a comprehensive data security solution is essential for modern enterprises coping with vast amounts of sensitive information. The success of our solution demonstrates that a hybrid approach combining deterministic and probabilistic methods, such as pattern matching and advanced AI/ML models, is effective in identifying and flagging sensitive data with overly permissive access controls. Our approach underlines the importance of a sensitivity rules database, categorized into Core, Common, and Unique Rules, to tailor data protection strategies to the specific needs of an organization. Join us in this session to learn more about this extensible framework that organizations can implement to secure their sensitive data in an increasingly complex digital environment.
Speaker
Maryam Nozari
Senior Data Scientist, Enterprise Knowledge LLC
Urmi Majumder
Principal Consultant, Enterprise Knowledge LLC
