NSF Award
2431614
The United States (U.S.) transportation sector remains a cornerstone of the economy, contributing over 8% to the country's Gross Domestic Product (GDP). Electrification efforts are transforming this sector, aiming to enhance mobility efficiency, reduce operating and maintenance costs, and cut greenhouse gas emissions. These efforts also seek to boost energy independence and security while significantly contributing to employment, particularly in technology and innovation fields. This shift has already placed more than 2.5 million Electric Vehicles (EVs) on U.S. roads, supported by over 70 thousand charging stations nationwide. To manage this advanced and complex cyberinfrastructure (CI), EV operators and vendors rely on cloud-based EV Management Stations (EVMS), crucial for provisioning services such as charging, billing, and authentication. However, the critical nature of EVMS has made them targets for malicious attacks, often state-sponsored, exploiting rarely investigated vulnerabilities. In response, this project establishes a collaborative ecosystem among academia, industry, and the public sector to bolster the resilience of the EV CI. It aims to develop proactive methodologies to identify and analyze Internet-connected EVMS and their software, thoroughly exploring and mitigating related vulnerabilities. This initiative connects several diverse Minority Serving Institutions (MSIs) within the established ecosystem, fostering joint research and providing enriching training opportunities. Through workshops, capstones, curricula material, virtual hands-on labs, professional development, and mentorship programs, the project enhances cross-disciplinary capacities at MSIs and beyond, driving forward the future of resilient, electrified transportation. In this context, this project serves NSF's mission in promoting the progress of science and securing national defense related to this ever-evolving CI.<br/> <br/> <br/>The project pioneers advanced fingerprinting techniques employing automated web scraping, recursive unsupervised learning algorithms, and pattern matching methodologies to identify and cluster Internet-scale EVMS. The primary objective is to detect deployed configurations and their interconnections, while retrieving critical artifacts, such as firmware binaries and compiled software, for comprehensive vulnerability analysis and disclosure. Leveraging robust industry connections, the project acquires auxiliary artifacts, including EVMS source code, through advanced supply chain reconnaissance and reverse engineering methods. This initiative also devises and implements an advanced digital forensic methodology rooted in ensemble techniques and machine learning classifiers. It integrates static analysis, file system forensics, memory forensics using volatility frameworks, data carving with custom heuristics, offensive security tactics, behavioral analysis through dynamic instrumentation, and virtualization methodologies such as hypervisor introspection to meticulously analyze the security posture of EVMS firmware and web endpoints. Furthermore, the project exploits state-of-the-art innovations in Large Language Models (LLMs) to automatically identify vulnerabilities in EVMS source code and suggest tailored and sound code fixes. This is accomplished by creating an unprecedented instruction-based training dataset using supervised fine-tuning, reinforcement learning, and transfer learning techniques. Additionally, the project establishes a large-scale data and threat repository to index discovered threat models, associated vulnerabilities, and retrieved EVMS artifacts. Accessible via RESTful APIs and web-based interfaces, this repository democratizes knowledge by making the harvested EVMS assets available at large, significantly empowering EVMS-centric threat situational awareness while fostering advanced research and development.<br/><br/>This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.
