Project Grant 2453331

Federal Grant Award Summary

The National Science Foundation's Division of Computer and Network Systems awarded $450,000 under the Computer and Information Science and Engineering (CISE) program (CFDA 47.070) to New Jersey Institute of Technology on May 15, 2026, for a project titled "Improving the Security of Large Language Model-Assisted Coding." Scheduled for completion by April 30, 2029, this research initiative will deliver methodologies and tools to enhance the security of code generated by Large Language Models (LLMs). The project will produce three primary deliverables: (1) program analysis techniques that bridge security properties and LLM code generation through contextual guardrails and prompt engineering; (2) an iterative LLM-centered code generation approach with security-focused criteria for successive improvement; and (3) a minimization approach to produce minimal code examples when LLM generation fails or cannot converge toward secure solutions.

The research outcomes will advance the practical application of LLMs in software development by empowering programmers to understand and mitigate security risks inherent in automatically generated code. The technologies developed will be applicable across multiple scenarios requiring formal verification, specification compliance, and robust code generation practices, thereby contributing to improved national cybersecurity infrastructure and economic productivity in the software development sector.