Artificial intelligence has become the opening topic of almost every anti-money laundering (AML) discussion this year.
According to RegTech firm Napier AI, that enthusiasm risks pushing compliance teams in the wrong direction. Vendors are promoting it, roadmaps are built around it, and institutions feel pressure to adopt it quickly. Napier AI argues that AI is not a shortcut to effective compliance and can only perform as well as the foundations it is built on.
Napier AI says the core problem is that many programmes start with the technology instead of the problem it should solve. Firms often add a new model to an existing process to reduce alert volumes or speed up investigations, without first checking whether the underlying controls work. If those controls are flawed, AI will not fix them. It will make the same weaknesses harder to detect.
The company also challenges a common assumption: that AI’s main purpose is to eliminate false positives. Napier AI points out that an AML system producing no alerts at all is not efficient. It is more likely to be absorbing risk it should not. The better question is whether a system generates the right alerts, and whether firms can prioritise, review and evidence them in line with their risk appetite. Regulators including the UK’s Financial Conduct Authority (FCA) and the Monetary Authority of Singapore (MAS) consistently expect firms to understand their risk and apply proportionate controls. In Napier AI’s view, AI should support quicker and more consistent decisions within that framework, not replace the judgement that shapes it.
Risk appetite therefore sits at the heart of Napier AI’s argument. Deciding which customers, products and geographies matter most is a compliance call, and it must be made before AI is introduced. When institutions set thresholds that reflect genuine exposure and reduce noise at the source, AI can focus on its strengths: finding complex patterns and helping analysts move faster. Without that groundwork, AI simply becomes a more sophisticated way to run an inefficient process.
Explainability is another non-negotiable, according to Napier AI. In many sectors an AI tool only has to be accurate. In AML it has to show its reasoning. If a model cannot explain why an alert was closed, escalated or deprioritised, firms have a governance problem rather than a productivity gain. Analysts need to trust the outputs, auditors need a clear trail, and regulators increasingly demand transparency, with the EU AI Act the most prominent recent example.
Napier AI also cautions against AI tools that sit outside the core AML platform. Such add-ons can look like quick wins, but they often force firms to govern two systems and lose context at each handoff. Embedding AI keeps recommendations, feedback loops and audit trails within a single operating model, making it easier to monitor, tune and explain over time.
Finally, the company stresses the role of the human analyst. The strongest deployments make people more effective rather than less involved, giving reviewers the authority, time and understanding to override recommendations. As agentic and more autonomous capabilities emerge, that oversight becomes even more important.
Napier AI concludes that the winners will not be the earliest adopters but the firms that build solid foundations in risk appetite, governance, explainability, workflow and architecture before scaling up. To support this, it has developed an AI Readiness Checklist, set out in its white paper, AI foundations for AML: building compliance-first intelligence.
For more, read the full story here
Copyright © 2026 FinTech Global
Copyright © 2026 RegTech Analyst
