STRP

New Jersey Institute of Technology

XCopilot: Private Code Generation with Large Language Models

Lead: Hai Phan

Partner: Ruoming Jin (Kent State University), Jonathan Maletic (Kent State University), Yelong Shen (Microsoft Azure AI)

Amount: $25000

Intellectual Property Status: Patent

Award Date: 05/31/2024

End Date: 06/01/2025

ABSTRACT

The demand from both individuals and enterprises for integrating commercial-grade AI-powered (software) code generation tools into their software development pipelines has significantly increased. This emergent market is projected to reach $264 billion by 2032 across a wide range of industry sectors, i.e., healthcare, retail, education, manufacturing, etc. However, many individuals and enterprises are afraid to use these tools due to potential security and privacy risks (i.e., OpenAI or Microsoft learn the IP of third-party companies that use their AI copilots). The project aims to develop XCopilot, the first privacy-preserving system that protects users’ prompts and generated software codes from being observed by AI code generation service providers (e.g., ChatGPT, Microsoft Copilot). XCopilot is secure, cost-effective, and deployable in industrial settings. As a result, it will revolutionize existing AI copilot systems by addressing their security and intellectual property (IP) leaking risks.