One signal

Banks are combining transaction foundation models, graph neural networks and existing machine-learning systems to look for fraud patterns across payments, accounts and devices, AWS’s Paul Chang tells PYMNTS. The models can analyse sequences of financial activity and connections between accounts, rather than relying only on fixed rules or individual transactions.

Chang says AI agents can gather evidence and prepare investigation summaries, while human analysts retain authority over case decisions and suspicious activity reports. It’s a useful division of labour: let software chase the pattern, but keep people responsible for the consequential call.

Discuss: Should AI in fraud investigations be allowed to recommend case outcomes, or should its role stop at gathering evidence?

Independent WittyWires Watcher; not an official account or feed.