LO
DeploymentCognitive automation2024-12-01
US Treasury reported that firms told it AI is widely used for AML and sanctions compliance, and that generative AI has been deployed to automate report creation and filing
Compliance officeroccupation page →
Event date / reported
2024-12-01
Evidence stage
DeploymentAn employer has put it into production. Can move the baseline — weighted by scale and how similar the setting is.
Tasks this bears on
Working the alerts
Transaction monitoring, sanctions screening, the flagged payment — deciding which of hundreds is worth investigating.
Automating✓ Evidence-backed
Proving it was done
Keeping the record that shows a control operated, a decision was reviewed and a policy was followed — for a regulator who will arrive later.
Automating✓ Evidence-backed
Where this applies
US financial services, published December 2024, synthesising comment letters submitted to a Treasury request for information issued in June 2024 and closed that August. Read what kind of evidence this is: it is a government department reporting what respondents told it, not a survey with a sampling frame and not a count of installations — anyone can write a comment letter and the firms that do are the ones with something to say. What makes it worth recording anyway is that it is the authority of record publishing under its own name, that the uses named are specific rather than aspirational (anomaly detection, flagging suspicious activity, identity verification under Bank Secrecy Act obligations, and collating and summarising data inside an investigation platform), and that report creation and filing — the evidence half of this job — is named as already deployed rather than planned.
What this means
The department that writes the rules published what the regulated firms told it they are doing, and the uses it names are the two tasks this page identifies as exposed: the alert queue and the evidence file. Report creation and filing is named as deployed rather than planned, which is the specific claim worth holding onto — it is the half of compliance work that produces a document rather than a judgement.
What it does not yet show
This is a synthesis of comment letters, not a survey: there is no sampling frame, and the firms that write in are the ones with something to say. It establishes that these uses exist and are described by the industry as widespread; it does not establish how many institutions, how much of the work, or whether any compliance post changed. Treasury is reporting, not measuring.
What you can check
Ask whoever owns your transaction monitoring what share of alerts were closed without a human opening the case last quarter. If nobody can answer, that is the finding, because the regulator's expectation is that alerts are investigated rather than closed.
Does it change the assessment?
No. The impact index is never moved by a single event. What this record did: the 2 linked task judgements above now rest on evidence instead of inference.
Source
U.S. Department of the Treasury, Artificial Intelligence in Financial Services (report on the June 2024 RFI) · verified 2026-09-13 · Claude (VOLO agent) — Treasury PDF downloaded and the AML/CFT passage read verbatim; the December 2024 date and the June 2024 RFI origin taken from the document itself · interpreted 2026-09-13 · Claude (VOLO agent)
Primary source — published by the party that did this, or the authority of record. No co-signature needed.