Sardine launches AI research lab and offers $375,000 fellowships to advance foundation models in financial crime detection, promising to improve fraud detection accuracy and address industry challenges.
Sardine has set up a new applied research arm, Sardine AI Labs, as it looks to push frontier artificial intelligence deeper into fraud and financial crime prevention. Alongside the lab, the company is offering $375,000 in research fellowships to independent researchers, a sign that it wants outside expertise to help turn its ideas into production tools for banks and fintechs.
The launch reflects a broader bet that financial activity can be read by foundation models in much the same way that language models interpret text. Sardine says patterns across transactions, devices, identities and internet addresses can be combined into behavioural signals that help systems detect anomalies, identify emerging attack methods and support risk teams making decisions under pressure. On Sardine’s own description of the lab, the work will focus on sequential modelling, transfer learning, adversarial robustness and explainability, with the aim of spotting attacks the model was never explicitly trained to see.
That research agenda speaks to a stubborn problem for financial institutions. Fraud prevention, sanctions screening and anti-money-laundering checks all add cost and delay to card, ACH and wire payments, while false positives can leave legitimate customers stuck in manual review for days or even weeks. The company argues that existing tools often miss some illicit activity while mistakenly blocking good transactions, exposing institutions to both regulatory scrutiny and avoidable losses.
Sardine says its early work has already produced promising results. In a white paper and related blog post, the company described a transformer-based foundation model trained on card transaction sequences without fraud labels, using roughly one billion transactions from more than a dozen card issuers over two years. Sardine says the approach lifted fraud detection performance and helped address the cold-start problem, in which newer issuers have too little history for traditional systems to perform reliably. In tests on issuers excluded from training, the company said the model improved detection accuracy by 68% for a consumer card issuer and 41% for a business card issuer compared with conventional machine learning methods.
Soups Ranjan, Sardine’s chief executive and co-founder, said the company is unusually well placed to build this kind of system because of the scale of its data network. He said, “We are uniquely positioned to train a model purpose-built for risk because we sit on the industry’s fastest growing fraud and fincrime dataset, which spans more than 6.5B devices, 441M consumers, 3.4M businesses, 6.6 billion transactions and $1.8 trillion in payments. Sardine AI Labs will focus on production-grade models that meet real world latency, governance, and explainability requirements.” Niranjan Shetty, Sardine’s head of data science, said the most important finding was that foundation models can learn directly from transaction behaviour rather than from hand-built tabular features, and that the next challenge is making those models fast, reliable and transparent enough for real-world risk decisions.
Ranjan also framed the effort as part of a wider shift in artificial intelligence, comparing Sardine’s approach with the multi-modal systems emerging from the leading AI labs. In his view, the next breakthrough in fraud and financial crime prevention will come from models that can work across device, identity, behavioural and transaction data at once. Sardine’s fellowship programme suggests it hopes to accelerate that effort by bringing in independent researchers to test whether specialist, domain-rich models can outpace the fraud rings they are built to stop.
Disclaimer: This article is intended to inform and educate, not to recommend or endorse any financial product, investment or strategy. Please consider your own financial circumstances and seek professional advice where appropriate before making financial decisions.
