\n <\/p>\n
![\"Recommended](\"https:\/\/www.europesays.com\/ai\/wp-content\/uploads\/2026\/05\/a71ab4ae-a627-4a6e-ada3-f6b8ae73de1d.png\")
<\/p>\n
The \u2018uncomfortable fiction\u2019 of AI agent compliance<\/p>\n
<\/a><\/p>\n
Financial crime compliance can sometimes be discussed as if the danger sits only in the new technology: the model hallucinating, the system missing a sanctions hit, the bank being unable to explain an automated conclusion to supervisors. Baxter\u2019s starting point is different. His question is what risks already exist in the current operating model \u2014 and whether the new system provides a better handle on them.<\/p>\n
\u201cMy initial question is, at what cost?\u201d he says of the promised productivity improvement. \u201cWhat\u2019s the risk that I\u2019m taking in order to do that?\u201d<\/p>\n
The answer, in Baxter\u2019s view, lies partly in the nature of the work being handed to the agents. These are not open-ended decisions about whether to bank a customer or file a suspicious activity report. They are structured tasks that already sit inside established BSA\/AML and sanctions processes.<\/p>\n
In enhanced due diligence, Dawson says, analysts previously had to gather and review transaction histories, check previous alerts and cases, and summarise relevant information before they could move into higher-value analysis. The digital EDD analyst now performs much of that foundational review and produces an auditable summary for an investigator.<\/p>\n
\u201cInstead of starting at A,\u201d Dawson says, \u201cyou\u2019re starting at F.\u201d<\/p>\n
The sanctions example is more straightforward still.\u00a0<\/p>\n
Chuck Taylor, Nasdaq Verafin\u2019s head of AML commercial strategy, gives the example of a customer whose name resembles that of someone on a sanctions list. A human analyst would compare identifiers such as age, country of origin and passport number, then document why the alert was a true match or false positive. The digital sanctions analyst performs that same comparison and produces a written rationale.<\/p>\n
The consistency of that work product is also part of the attraction. Baxter points out that human analysts do not document identical reviews in identical ways. That variation may be manageable, but it takes effort to validate.<\/p>\n
\u201cYou\u2019ve got 30 humans who are documenting a process and their work,\u201d he says. \u201cThere is an inherent inconsistency in the way that maybe the five of us documented doing exactly the same project, and that inconsistency takes time to validate. We now have a consistent output that we can validate against.\u201d<\/p>\n
Taylor says that consistency also matters to supervisors. Regulators reviewing sanctions alerts or enhanced due diligence files may question why similar cases were handled or documented differently by different analysts. With the agents, he says, \u201cit becomes much more consistent and defendable because you\u2019re doing it exactly the same way\u201d.<\/p>\n
Baxter goes further. In a human-led process, quality assurance may involve periodic samples of analyst output. With the agents, he argues, the bank can monitor the full population of work.<\/p>\n
\u201cI can monitor these agents effectively in real time,\u201d he says. \u201cI can monitor every action that agent takes, in total, and I can do it continuously.\u201d<\/p>\n
Baxter is also clear that the bank\u2019s comfort depends on its own institutional capabilities. FNBO is not a large Wall Street bank, but Baxter describes it as \u201csomewhat of a unicorn\u201d. Its national credit card business has made it a data-heavy institution for decades, while its community banking footprint gives it a more traditional relationship business.<\/p>\n
\u201cFor a $34bn bank, our data sciences group is pretty damn good,\u201d he says. \u201cWe\u2019ve been playing around with AI, machine learning, all your heavy data work, for probably 10 or 15 years.\u201d<\/p>\n
Comfortable with regulators<\/p>\n
That background matters because Baxter sees risk appetite as inseparable from knowledge.<\/p>\n
\u201cWe get comfortable with it quickly, and we have a risk infrastructure designed to enable us to use this kind of technology that meets with our appetite for risk.\u201d<\/p>\n
It also shapes how the bank communicates externally. Baxter says FNBO has developed \u201ca really good way of working with regulators\u201d and does not treat supervisory engagement as a box to be ticked after a project is complete.<\/p>\n
\u201cWe are incredibly transparent with our prudential regulator,\u201d he says. \u201cIf they\u2019ve got a job to do, how do you expect them to be able to do that job when you keep them in the dark and only bring them in when they have to be brought in?\u201d<\/p>\n
We tend to incorporate and involve [regulators] relatively early on in these processes, so they get comfortable as we get comfortable<\/p>\n
On agentic AI, he says, the bank has tried to bring regulators along as its own understanding develops.<\/p>\n
\u201cWe tend to incorporate and involve them relatively early on in these processes, so they get comfortable as we get comfortable,\u201d Baxter says.<\/p>\n
Baxter\u2019s view is that the conversation becomes easier if the bank can show that the use case is bounded, the outputs are auditable and the controls are stronger than the process being improved.<\/p>\n
The same logic applies to law enforcement. Baxter says the aim is not merely to push suspicious customers out of FNBO and into another bank. He wants investigators to spend more time building cases that police and prosecutors will actually act on.<\/p>\n
\u201cThis allows us to focus on building cases that\u2009.\u2009.\u2009.\u2009are attractive enough that they will then pursue [and get] people put in jail or taken out of the system,\u201d he says.<\/p>\n
Asked whether the bank has already had material law enforcement success since using agentic AI, Baxter says: \u201cNo, things don\u2019t move quite that quickly.\u201d But the strategic ambition is clear. If AI can take on the repetitive work of gathering and structuring information, investigators can spend more time on the judgment-heavy work that makes a case useful.<\/p>\n
Next steps<\/p>\n
That is also why Baxter is interested in taking the approach further. FNBO is looking at other agentic AI projects, including regulatory compliance, complaints management and data extraction. But he frames the opportunity in incremental rather than revolutionary terms.<\/p>\n
\u201cWe\u2019re not looking for a moonshot,\u201d he says. \u201cWhat we\u2019re doing is we\u2019re getting really good at some very well tested, very reliable implementations of this technology, and scaling them horizontally across processes.\u201d<\/p>\n
For a banking audience, that may be the real lesson. If an AI agent can produce a more standardised file, expose its steps to monitoring and leave humans to make the more complex judgments, then the risk calculation starts to look different.<\/p>\n","protected":false},"excerpt":{"rendered":"First National Bank of Omaha has put agentic AI to work in sanctions screening and enhancing due diligence,…\n","protected":false},"author":2,"featured_media":38439,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[6],"tags":[179,7493,23790,7145,1715],"class_list":{"0":"post-38438","1":"post","2":"type-post","3":"status-publish","4":"format-standard","5":"has-post-thumbnail","7":"category-agentic-ai","8":"tag-agentic-ai","9":"tag-agentic-artificial-intelligence","10":"tag-crime-conduct","11":"tag-operations","12":"tag-regulation"},"_links":{"self":[{"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/posts\/38438","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/comments?post=38438"}],"version-history":[{"count":0,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/posts\/38438\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/media\/38439"}],"wp:attachment":[{"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/media?parent=38438"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/categories?post=38438"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/tags?post=38438"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}