Napier AI has weighed in on one of the thorniest debates in financial crime compliance: the gap between what artificial intelligence can theoretically do for transaction monitoring (TM) and what institutions are actually achieving today.
The discussion, raised during a recent ACAMS New Jersey chapter session, centred on a widening divide between AI used for efficiency and AI used for genuine detection.
Napier AI’s Michael Joseph noted that most AI in TM today is used for efficiency, helping investigators clear alert queues or draft the first version of a Suspicious Activity Report (SAR). That work has value, but the detection itself has already happened via rules. A smaller group is now putting AI on detection directly, particularly since the US Department of the Treasury’s Financial Crimes Enforcement Network (FinCEN) issued a Notice of Proposed Rulemaking (NPRM) in April, reforming AML/CFT programmes under the Bank Secrecy Act.
FinCEN has named AI directly as a factor in enforcement decisions, shifting the industry question from how much time AI saves to how much crime it actually helps identify. Napier AI’s Michael Joseph described this as a welcome, if still evolving, shift towards outcomes over check-the-box compliance.
On SARs, the piece argues against “super SARs” overloaded with detail, calling instead for precision, with AI-generated audit trails supporting clearer narratives for law enforcement and compliance teams alike.
Napier AI also cautions against mistaking speed for progress. If false positive rates remain above 90%, faster processing of the same alerts does not equate to reduced risk exposure. Alert volumes typically rise in year one of AI-augmented monitoring, not fall, as models surface previously unhandled categories of activity.
On automation limits, Napier AI maintains that AI should not auto-close alerts or file SARs end-to-end for the foreseeable future, since accountability for financial crime decisions cannot be outsourced to a model. Instead, the better fix is tuning detection so fewer low-value alerts are generated in the first place.
Ultimately, Napier AI frames rules and AI as complementary rather than competing: models surface what rules cannot, and once a pattern is validated through real SARs, it gets folded back into an explainable rule, creating a feedback loop regulators are more likely to trust.
For more insights, read the full story here.
Copyright © 2026 FinTech Global
Copyright © 2018 RegTech Analyst
