Banks running anti-money-laundering and fraud checks on large language models keep hitting the same wall: a system that flags a transaction but cannot show its reasoning is difficult to defend to an examiner. Rabobank’s renewed AI partnership with Expert.ai points to how the industry is answering that problem, not with a bigger model, but with an architecture built to be audited.
On July 15, Rabobank and Expert.ai announced an extension of a partnership that dates to 2018, widening the Dutch bank’s use of Expert.ai’s EidenAI Suite across customer screening, financial crime detection, anti-money-laundering monitoring, and document processing. The suite pairs natural language understanding and knowledge graphs with machine learning and large language models inside what Expert.ai calls a “fully governable, multi-agent framework,” rather than a single end-to-end LLM pipeline. “Rabobank has consistently demonstrated leadership in innovation and a forward-thinking vision for the role of AI in banking and financial services,” said Andrea Ricotti, Expert.ai’s SVP Sales North Europe, in the announcement.
For compliance officers, that architecture choice is now a regulatory one. AML and financial crime tools sit inside exactly the kind of decisions supervisors want explained, and a pure LLM produces a probability score, not a traceable reason. Layering knowledge graphs and rules-based language processing underneath the model gives banks something closer to a paper trail: a flag can be tied to a specific extracted entity or rule, not just a confidence number.
The eight-year length of the relationship is itself informative. As US regulators shift AML supervision toward a risk-based standard, the AI vendors best positioned to win renewal are not the ones with the newest foundation model, but the ones with years of production history a bank can point to when an examiner asks why the system should be trusted.
Source: Expert.ai
