AML False Positive Rate Statistics: 2026 Report

Published: November 9, 2026By Benjamin Margolis, Founder and Managing Editor, The Fintech Index

Our analysts reviewed published customer results from the AML vendors we rate, FinCEN's annual filing statistics, and Unit21's data on AI alert review, as of Q4 2026. We compared reported false-positive reductions, tracked suspicious activity report volumes by fiscal year, and measured how often AI closes alerts without a person.

Vendor case studies describe one customer each and use different measures, so every row keeps its own metric.

Rated Vendors Report False-Positive Cuts of 50% to More Than 90%

We collected the customer results that rated AML vendors publish on their own sites. Each case covers one customer, and vendors choose which results to publish, so the rows show what a well-run project can reach, not an industry average.

Key Findings

• Screening projects report the largest cuts, led by Banco do Brasil at more than 90% and AJ Bell at 82%. Fuzzy matching against sanctions and PEP lists creates many near-identical hits, so better matching has more to remove.
• Transaction monitoring results cluster lower in this set, at 50% to 60%. Ebury nearly doubled the share of actionable alerts, from 14% to 30%, which still leaves 70% of its alerts without action.
• LHV's cut from 1% to 0.25% of transactions sent to manual review let a single agent handle more than 13,000 alerts a month.

US SAR Filings Rose to 4.8 Million in FY2025

FinCEN publishes the number of Bank Secrecy Act reports filed each fiscal year. A suspicious activity report is the end point of an alert that survives review, so SAR volume shows how much monitoring output reaches the regulator.

Key Findings

• SAR volume grew about 12% from FY2022 to FY2025, and most of that growth came in FY2023. Annual growth has held near 2% since then.
• Depository institutions filed 2.8 million SARs in FY2025 and money services businesses 1.26 million. Securities and futures firms filed about 102,000, casinos about 77,000, and loan or finance companies about 44,000.
• About 335,000 registered institutions and other filers submit BSA reports, an average of about 14 SARs per filer in FY2025 that hides wide differences between large and small filers.

AI Closes About 0.5% of Alerts on Its Own

Unit21 published usage data from about a year of live AI alert reviews, covering more than one million alerts across its customers. It splits AI use into three levels, from research support to fully automatic closing.

Key Findings

• Most teams keep a person in the decision. Assist mode accounts for about 84% of AI usage and auto-close for about half a percent.
• AI-worked alerts found true positives about twice as often as human-only alerts overall, about 15 times as often in fraud workflows, and about 7 times as often in enhanced customer review (EDD) cases.
• One organization's simple check-deposit threshold ran at a false-positive rate of about 99%, the kind of queue where analyst fatigue leads to missed fraud.

False Positives Still Dominate AML Alert Queues

False positives remain the largest cost in AML operations. Published vendor results show that tuning and AI can remove half or more of them, while SAR filings keep rising by about 2% a year, so alert queues are not getting smaller on their own.

Looking ahead, the share of alerts that AI closes without a person should rise from today's half a percent as examiners grow comfortable with agent decisions and audit trails. FinCEN's work to modernize the AML regime could also change which activity institutions must report.

So if you run an AML alert queue, measure your own false-positive rate and true-positive yield first, then ask each vendor to repeat its published result on a sample of your data.