Artificial intelligence has sharpened detection, cut false positives and surfaced patterns that rules-based anti-money laundering (AML) systems simply cannot see.
Yet according to new analysis from Napier AI, many institutions remain stuck running pilots rather than scaling AI across their compliance operations, held back not by technology but by trust.
Napier AI’s research argues that the real barrier to adoption is confidence. As regulators build explainability, auditability and governance into supervisory expectations, trust is becoming the deciding factor in whether AI moves from experiment to core infrastructure. Firms must now show not just that a flag was raised, but why, what data drove the decision, and how it aligns with regulatory expectations.
This marks a shift from process-based compliance towards outcomes-focused scrutiny, Napier AI notes, with regulators examining both accuracy and the reasoning behind it.
The UK’s Financial Conduct Authority has leaned into this through programmes such as its Supercharged Sandbox and AI Live Testing, allowing firms to trial models under controlled, closely governed conditions. The message is consistent: innovation is welcome, but only where it can be understood.
Central to this is the move from “black box” systems to “glass box” transparency, embedding explainability into AI by design rather than bolting it on afterwards. Napier AI stresses that glass box models must give investigators and auditors clear, contextual reasoning at the point of decision-making, not a reconstruction after the fact.
Even well-performing large language models and agents can generate flawed reasoning behind correct alerts if poorly designed, meaning explanations must be regularly spot-checked rather than trusted blindly.
Globally, momentum is building. Beyond the FCA’s collaborative sandbox approach, the EU’s new Anti-Money Laundering Authority has launched an industry-wide data collection exercise to test risk assessment models ahead of direct supervision from 2028. Napier AI observes that regulators are increasingly acting as facilitators of responsible innovation rather than pure overseers, though institutions in less clearly defined markets still face structural, rather than technological, uncertainty.
Napier AI’s own AML Index 25-26 found that heavier compliance spend alone has not delivered greater effectiveness. AI could unlock roughly $183bn a year in global savings, but only where models are explainable, decisions auditable and human oversight preserved. A human-in-the-loop approach keeps compliance professionals accountable for outcomes, with AI enhancing rather than replacing judgement.
For Napier AI, the conclusion is clear: glass box AI is not a future aspiration but an immediate requirement. Scaling AI in AML is no longer about proving it works, but proving continuously that it can be trusted.
For more, read the full story here.
Copyright © 2026 FinTech Global
