Sardine launches AI research lab to fight financial crime

What happened

Sardine, a platform specialising in agentic risk management for fraud and financial crime prevention, has launched Sardine AI Labs, a new applied research unit dedicated to advancing frontier AI techniques for combating financial crime.

According to Finextra, the lab will focus on research and development work aimed at pushing the boundaries of how artificial intelligence can be deployed against fraud, scams and other forms of financial crime, building on Sardine's existing risk platform.

Why it matters

Financial crime techniques are evolving quickly, particularly as fraudsters themselves begin to use generative and agentic AI to scale scams. A dedicated research lab signals that Sardine intends to compete on the frontier of detection capability rather than relying solely on incremental updates to existing models, positioning applied research as a core part of its product roadmap rather than a peripheral function.

For banks, fintechs and payment platforms, this points to a broader shift: risk and compliance functions are increasingly becoming testbeds for advanced AI, not just back-office cost centres. How quickly defensive AI capability can match offensive misuse of AI will shape trust and loss rates across digital financial services.

The Renascence take

Fraud prevention is ultimately an experience problem as much as a technical one — every false positive is a frustrated customer, and every false negative is a breach of trust that's hard to repair.

The real test for Sardine AI Labs won't be how sophisticated its models become, but whether that sophistication translates into fewer legitimate customers getting blocked, delayed or wrongly flagged. Fraud teams often optimise for catch rates while losing sight of the friction imposed on honest users; a research lab gives Sardine room to tune that trade-off more deliberately rather than defaulting to blunt risk thresholds. Operators evaluating this kind of capability should ask vendors not just "how much fraud do you stop" but "how little disruption does it cause to everyone else."