Consumer card issuers saw fraud detection accuracy rise by 68% during testing of a new foundation model for financial crime, while business card issuers saw a 41% increase. The system was developed by Sardine AI Labs, an applied research group that leverages data from 6.5B profiled devices and $1.8T in payments to identify fraud for card programs that lack their own historical datasets.

The company is also establishing the Sardine AI Fellowship Program, providing $375,000 in total funding for up to five researchers and faculty members from US and Canadian universities. Fellows will work with internal researchers on identity matching, scaling laws, and real-time inference to improve how AI models generalize across different financial institutions.