CICI:RSSD: Live Evaluations of Real-World Security Data Lake from National Cyberinfrastructure

About This Grant

Artificial Intelligence(AI)-driven cyberattack detection is essential for safeguarding the U.S. supercomputing infrastructure. Research in AI relies on the national supercomputing infrastructure, but this critical resource is vulnerable to cyberattacks. Securing this infrastructure requires an extensive understanding of historical security incidents, providing a longitudinal perspective on trends, seasonality, and the evolution of cyberattacks. Without this historical context, the research community is left to react rather than preempt futuristic threats, such as AI-driven malware, quantum-resistant vulnerabilities, and machine learning model supply chain backdoor attacks, leaving scientific breakthroughs vulnerable. The AICyberLake project curates a security data lake by sourcing cyberattacks from the DeltaAI system at the National Center for Supercomputing Applications (NCSA) and its peer supercomputing centers. The data includes Zeek network cryptographic metadata, graphics processing unit (GPU) interconnect vulnerabilities, and ground truth incident reports. The resulting data lake provides a real-time, anonymized stream of attack attempts to vetted research teams for evaluating their agentic AI-based detection models against unseen adversaries. An API (Application Programming Interface) helps inform the broader community by contributing attack metadata to policymakers such as the National Institute of Standards and Technology (NIST) and the Cybersecurity and Infrastructure Security Agency (CISA). Ultimately, the project aims to reinforce public trust in running AI workloads within cyberinfrastructure, provide practical exercises by reproducing novel attack traces, and helping educate the next generation of the cybersecurity AI research workforce. 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.