CERTIFY: Contractual Integrity for Scientific AI-Ready Data Pipelines is funded by the NSF’s Cybersecurity Innovation for Cyberinfrastructure (CICI) program. Charles Cao of the University of Tennessee, Knoxville leads the three-year, $900,000 project, with Tabitha Samuel and Gary Rogers of NICS as co-principal investigators.

Scientific data now passes through long pipelines, often spanning several institutions, before it reaches an AI model, and corruption or an undocumented change at any stage can compromise every result built on it. CERTIFY develops a framework that lets datasets carry machine-checkable proof that specified integrity and provenance requirements have been met, so collaborating teams can trust shared data without re-auditing one another’s work. The approach pairs temporal contracts with operating-system-level monitoring and AI agents to detect and contain problems as data moves through multi-institutional pipelines. The team will evaluate the framework on agricultural sensor networks and electronic health records, and will release open-source software, documented datasets, course modules, and training materials to strengthen reproducibility in data-intensive science.

CERTIFY is supported by NSF award 2613350.