UBS’s brokerage arm is paying $125 million for a monitoring promise it did not keep
FinCEN’s August 2026 penalty, the largest it has put on a broker-dealer, says more than 50,000 foreign-currency wires went unwatched after a 2018 order the firm said it would fix.
The second time is the point
On 3 August 2026 FinCEN assessed a civil money penalty of $125 million against UBS Financial Services Inc. for wilful violations of the Bank Secrecy Act. The agency called it the largest BSA penalty it has imposed on a broker-dealer. The size is the headline. The fact that it is the second FinCEN action against the firm is the case. In December 2018 UBS Financial Services entered a consent order and paid $14.5 million. That order found the firm had not adequately monitored foreign-currency wires because its automated system was weak. The firm told FinCEN it would fix the problem. On FinCEN’s 2026 account, it did not. The agency says the firm later failed to monitor more than 50,000 foreign-currency wires with an aggregate value above $10 billion, and did not tell FinCEN. Examiners found the gap. FinCEN learned of it in a later investigation, not from a voluntary disclosure. A promise inside a consent order is now exhibit one in a recidivism case. That is the sentence every firm that has ever told a regulator “the scenario is being rebuilt” should tape above the desk.
Wealth files failed in the same years
Customer due diligence failed in the same period, particularly for high-risk clients with ties to Russia and Latin America. FinCEN says the firm did not properly weigh source of wealth, or negative news alleging corruption ties. The order is about how wealth is taken on, not only about a broken alert. A file that collects a passport and a static risk score is the example the agency is warning against. Covered broker-dealers have to know high-risk customers at onboarding and again when the relationship changes. Foreign-currency wires are an obvious place to look. So is negative news. The order does not say every client with a Latin American or Russian tie is illicit. It says the work the law already required was not done. As part of the resolution the firm admits it wilfully violated the Act, including the duty to keep an anti-money-laundering programme that meets the minimum requirements and the duty to file suspicious-activity reports. A third party must look back for suspicious transactions the monitoring missed. An independent review will test the parts of the programme that produced the violations. Lookbacks are where the cost stops being the fine and becomes the archive. Anyone who prices a BSA penalty as the cheque alone has not priced a lookback.
Other regulators were already in the room
FinCEN noted cooperation with the Commodity Futures Trading Commission, the Securities and Exchange Commission, and FINRA. The same foreign-currency failure had already been the subject of a settlement involving FinCEN, the SEC, and FINRA. The 2026 order says meaningful remediation still took years. That is a useful correction to the idea that a joint settlement closes a subject. It closes the paper. The wires continue. If the scenario you promised to repair does not actually score them, the next agency in the room will not be impressed that you settled last time. For a committee that oversees outsourced transaction monitoring, the blunt question is this: if we promised a regulator we would repair a scenario, who checks that the repair scores the wires. Not who wrote the letter. Who sampled the output. Assurances in a response to a consent order are now a liability if they are untested. Vendor slides that say the model was enhanced are not a sample.
What $10 billion of unmonitored FX should change
Fifty thousand wires is not a rounding error and it is not a niche product. Foreign-currency wires are a core way wealth and brokerage clients move value across borders. An automated system that is weak, stays weak, and is not disclosed, is a programme failure even if every other scenario in the bank is pretty. Faircroft would ask any large broker-dealer, in 2026, for a one-page map of FX-wire monitoring: what is in the engine, what was promised to a regulator, what a sample of fifty wires showed last quarter. If the sample cannot be produced, the map is fiction. The same page should say how negative news is weighed for high-risk wealth clients, and how often. A static score assigned in year one is the file FinCEN described. Refresh is the control. If your wealth channel and your brokerage channel do not share that refresh, you have built the seam this order sits in.
What the penalty does not say
It does not say UBS the bank, as opposed to UBS Financial Services Inc. the broker-dealer, was fined this sum. Precision about the legal entity is the whole point of a correspondent file. It does not say every foreign-currency payment at every UBS entity was unmonitored. It does not invite gossip about individual Latin American or Russian clients who have not been named. It says this broker-dealer admitted wilful BSA violations after failing to do what a 2018 order required, at a scale of tens of thousands of wires and more than $10 billion, and that high-risk due diligence failed alongside the alerts.
The 2026 use of a recidivism case
Treat it as a template for any open promise to a regulator. List them. Date them. Sample them. If you cannot sample them, you do not have a promise, you have a sentence you hope nobody rereads. Independent reviews and lookbacks are slow. The decision to start one is not. A firm that waits to be told twice has now watched someone be told twice in public, at $14.5 million and then at $125 million. Faircroft would keep both numbers, both years, the wire count, the $10 billion, the admission, and the names of the other agencies. We would not smear the parent brand beyond the entity named. We would ask our own monitoring owner the question FinCEN has now made expensive: who checks that the fix actually scores the wires.
A lookback is a second book, not a paragraph
The 2026 resolution does not stop at the cheque. A third party has to look back for suspicious transactions the monitoring missed, and an independent review has to test the parts of the programme that produced the violations. That is months of archive work: pulling the unmonitored foreign-currency population, deciding what a miss looks like, and writing suspicious-activity reports that should have existed earlier. Firms that budget the penalty and forget the lookback will miss the real cost, and they will miss the reason the order is useful to everyone else. If your own open consent order has a “we will repair the scenario” sentence, the UBS chronology is the exhibit for why that sentence needs a sample attached. Fifty wires, scored, dated, and signed, beat another letter to the regulator that repeats the promise. In 2026 the only respectable status for a monitoring repair is: promised, built, sampled, and still scoring. Anything short of the sample is how $14.5 million becomes $125 million.