The choice of AI in financial crime compliance is no longer just about which model performs best. According to Napier AI, financial institutions are now weighing where a model was trained, who built it, what data it learned from and where it runs. Those factors don’t necessarily work against performance, but they are separate measures, and they are increasingly shaping AI strategy.

The theme was debated at a recent panel at the Napier AI / AML Assembly, which brought together practitioners from technology, financial services and payments. The session was held under the Chatham House Rule, so views were shared without attribution.

Panellists noted that running software in local or dedicated environments is nothing new. The harder part has been accessing certain AI models in a fully disconnected way, because some providers still don’t allow it. That restricts institutions with strict data residency or control requirements. The panel expected that as sovereign cloud offerings for AI mature, data control will become a baseline expectation rather than a premium extra.

The case for control goes beyond compliance. Napier AI pointed to a multinational corporate seeking to expand its credit facility, which typically approaches several banks with broadly the same information. Each bank could feed an identical prompt into the same general-purpose model. What separates them is the institutional and tacit knowledge they apply on the way to a decision. Training AI on that knowledge is a real differentiator. The same is true in financial crime compliance, where typologies, risk appetite and investigator judgement are hard-won assets.

Source link