Graph neural networks are turning hidden fraud into visible networks
SiliconANGLE Kelly Knight
Graph neural networks are helping Gilead spot fraud hiding across linked accounts and transactions. That matters because the bad actors weren’t visible in single-data checks.
Based on reporting by SiliconANGLE, Kelly Knight — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Graph neural networks are changing fraud work from a search for odd transactions into a hunt for entire hidden networks. At Gilead Sciences, that shift is showing up in anti-counterfeiting and commercial fraud detection, where the target isn’t one suspicious record but a web of connected entities.
Thomas Luu, Gilead’s director of global product security, said his team had to convert relational data into graph form to keep up with more sophisticated schemes. Speaking with theCUBE at Neo4j GraphTalk, he described a system built around three layers: rules for known patterns, traditional machine learning for statistical anomalies, and graph neural networks for the relationship-based fraud that the other two layers miss.
Before that, the work leaned heavily on manual comparison across separate and nuanced data sets. That approach depended on individual analysts and did not scale well. Once the underlying data was cleaned and unified, the team could automate more of the process and apply graph machine learning on trustworthy inputs.
The point of the graph layer is not just speed. It also helps surface the supporting cast around a main fraud actor, including lower-volume players whose activity might disappear inside aggregate data. And because the relationships are visible, the findings are easier to explain to investigators who are not technical specialists.
Luu’s line about fraudsters hiding in the averages gets at the whole thing. Traditional systems are good at counting events. Graphs are better at showing who is connected to whom, and in fraud, that’s often the part people are trying to bury.
My take — AI-written commentary, not fact-checked reporting
This is the rare AI story that sounds less like hype and more like plumbing finally catching up with reality. Fraud networks are social networks with worse morals, so of course the tool that maps relationships beats another shiny anomaly detector. The industry keeps selling intelligence as if it lives in single records; the money is usually in the links.
Read more about this at: SiliconANGLE
