Automated Software Test Generation and Augmentation for Improved Debloating

The DON26TZ01-NV008 RFP seeks to develop an automated solution for generating and augmenting software tests to improve proactive cyber techniques like debloating and software refactoring. The objective is to refine test suites that can serve as practical expressions of software transformation, leveraging multi-modal methods like ingesting code and documentation, and integrating with DevOps processes. Current software development, with its reliance on code reuse and AI, often leads to complex applications with numerous dependencies, many of which contain unneeded code that can introduce vulnerabilities. Existing debloating tools, which remove this excess code, often fail due to low-quality or incomplete debloat specifications, leading to broken applications. The proposed technology aims to address this by automating the capture of desired software behaviors, utilizing code analysis, documentation, build configurations, existing tests, and user input. Phase I involves developing a concept for multi-modal processing to generate and augment tests, with a focus on unstructured documents and user interaction. Phase II will prototype this capability, validating it in a containerized environment and demonstrating its ability to debloat applications using limited test suites, unstructured documents, and real-time user input. Phase III focuses on integrating this capability with Program of Record systems and developing commercial applications for automated software testing and fuzzing harness generation, particularly for AI-generated code.