- Key insight: First National Bank of Omaha has become a relatively early adopter of AI agents for financial crime investigations.
- What's at stake: The bank says it's cut investigation time in half; experts say there are risks banks need to keep in mind.
- Forward look: FNBO and others are looking to adopt AI agents for fraud detection work, among other use cases.
At First National Bank of Omaha, AI agents have begun taking on some of the work of human financial crime investigators, reducing the time that people spend on each case by 50%, according to bank executives. The agents are also helping human investigators respond consistently to fraud, money-laundering and sanctions-violation alerts.
The AI deployment is part of a broader trend in which U.S. banks are seeking a return on their investments in agentic AI. At the same time, financial crime is on the rise. Nasdaq Verafin's
"The consistent challenges of financial crimes means you're always looking for a better way to solve the problems," Nick Baxter, chief risk officer at FNBO, told American Banker.
Chuck Subrt, director of the fraud and AML practice at Datos Insights, said he sees the use of agentic AI in financial crime work as a rapidly emerging trend, though banks are still relatively early on the adoption curve.
"Across the financial services industry, agentic AI is gradually moving beyond pilots into in-production financial-crime operations," Subrt told American Banker. "Many of the fundamental elements of investigative groundwork, such as gathering transaction history, entity data and contextual signals and distilling them into a structured summary, are among the first areas where institutions are applying it. Agents are now completing many of the steps needed to complete the high-volume, repetitive evidence-gathering that is required to support the analysis and the decision, while the final judgment decision remains with the analysts and investigators."
What the AI agents do
FNBO was already using Nasdaq Verafin's anti-money-laundering and fraud analysis software. When the vendor showed the FNBO team the two AI agents it launched a year ago for enhanced due diligence and sanctions screening, the company was open to trying them.
"From Nasdaq's point of view, we've seen the need in the industry to help manage the rising complexity of regulations, as well as the rise in illicit financial activity," Rob Norris, senior vice president and head of product at Nasdaq Verafin, told American Banker. "There's an increase in the need to combat these bad actors, and our customers have been telling us across the board that they're looking for technology to help in that fight." The company's AI agents are developed using AWS Bedrock, an Amazon service for building generative AI applications, which according to Nasdaq Verafin allows them to be model-agnostic.
The enhanced due diligence AI agent does some of the work required when a bank receives an AML, fraud or sanctions-violation alert, to determine whether the alert truly indicates a financial crime has been committed that calls for a suspicious activity report to be filed, or whether the alert is a false positive.
The other AI agent helps banks research sanctions screening alerts. If a sanctions alert flags that a customer appears to be doing business with a person or company on the U.S. government's Office of Foreign Assets Control list, this AI agent can look at the customer's transaction history and pull in context that might prove or disprove the need to file a suspicious activity report.
"Our focus early on has been on what we call level-one alert triage," Norris said. "So you can think about all of the different types of alerts that systems generate for anti-money laundering or fraud, or even things like sanctions screening. Typically there's a level-one screening of what is a good alert versus a bad alert, what alerts deserve a deep dive versus those that can be explained away."
More than 650 banks use these agents, according to Nasdaq Verafin. The Agentic Sanctions Analyst has delivered up to a 90% reduction in sanctions alert review workload at some banks, according to the company, while the Agentic Enhanced Due Diligence Analyst has delivered up to a 50% reduction in time spent completing enhanced due diligence reviews.
At FNBO, the agents quickly research and document reasons why some alerts are false positives or represent a low risk, while others are serious and should be escalated to a human for review. One observable result has been better consistency, according to David Dawson, BSA/AML officer at FNBO.
"A lot of times there's a pretense that existing processes are already perfect, without fault or error," Dawson told American Banker. "We could try our best to use templates, but if you take 100 investigators or analysts, and you send them off each to do 100 enhanced due diligence reviews, you're going to have variability in how they document information." Using the AI agents, the foundational research is automated in a consistent way.
"I think of it like, if it's a 10-step process, in the past, we started on step one," Dawson said. "Today we're starting on step three, and it's exciting to have incremental progress through this type of innovation on those existing embedded processes."
The AI agents provide artifacts and summaries for human analysts to review, allowing the humans to redirect their time to the higher-value analysis, Dawson said.
For example, a sanctions alert might flag a transaction by a customer whose name is similar to someone on the OFAC list.
"But in research, maybe you can tell that it's a different jurisdiction, dates of birth or contact information," Dawson said. "It can say, this name is similar to this name, but they are clearly different people, and that information is used by the agents to rationalize and document for the bank why that is interesting, or it can be cleared."
The AI agents help the bank validate its financial crime investigations "at a far greater level of granularity than with the manual process," Baxter said. "And the big kicker at the end of the day is those human resources I freed up, I can now apply to more high-value work. So I've got productivity, I've got an improved work product, I've got a more effective monitoring of the work, and I've freed up resources I can put on higher-value work that's going to have a bigger impact on getting crooks off the street."
The new AI agents are not replacing people, Dawson said. "There is no shortage of financial crimes," he pointed out. The agents help human investigators "get bad actors in jail and protect the bank in our communities."
To preserve security and control access to systems, the AI agents are given the same access controls and privileges in the Nasdaq Verafin system that a human agent would receive for the same task.
"Before more access is granted for them to take more actions, you can preview how they would work, so you can think about all the different safety and governance measures that we would put in for a bank," Norris said. AI agents can be monitored more closely than humans, he added.
According to Dawson, FNBO's 15 financial crime investigators and analysts are excited about being freed from some of the foundational research so they can focus on higher-value analysis.
Dawson estimates that these AI agents are saving about 50% of financial crime investigators' time.
"If an [enhanced due diligence] case on average took a half an hour, that might take around 15 minutes now in comparison," he said. The AI agents are also helping the bank get to its true objective, which is being able to identify and detect crime and get it to law enforcement to be able to act. "This helps us balance blocking and tackling needs with our resources," Dawson said.
What's next
In the future, FNBO may use AI agents to research fraud cases — Nasdaq Verafin released an Agentic Fraud Analyst earlier this month.
"You start simple, you build up confidence, you validate your ability to monitor and manage these agents," Baxter said. "We started out relatively low risk with jobs that were relatively easy to explain with output that was consistent and allowed us to monitor it. That builds confidence of the operators, the monitors, the managers, the regulators, that we're moving with the right platform, and then we can get more adventurous."
Subrt offered a few cautions for banks that want to use AI agents to help investigate financial crime. For instance, for work related to sanctions screening and enhanced due diligence, any reduction in accuracy "carries direct regulatory consequence and requires very careful attention through the design, development and deployment as well as ongoing maintenance and scrutiny of the agent across its life," he said.
Explainability is important, he said. "The agent's work and summary must be traced to its underlying sources," Subrt said. "Otherwise, it introduces risk rather than reducing it."
Establishing and maintaining data integrity, quality and completeness are equally essential. "For example, if an agent omits a relevant signal or over-weights an immaterial one, it can create an erroneous or incomplete record, which may negatively influence the investigation work before an analyst or investigator starts their process," he said.
Institutions should also be wary of bias or hallucinations that could undermine the human judgment that the solution is intended to support. Financial crime organizations should be trained and encouraged to monitor and challenge agent output, not just blindly consume it, Subrt said.
Governance is a further consideration, he said. "Agents should operate within the institution's own policies, risk appetite and escalation thresholds, supported by clear audit trails and validation against historical cases before deployment, so performance is demonstrated rather than assumed," Subrt said. "Ultimately, these agents fall within existing model risk and third-party governance expectations, and the institution remains accountable for the outcome regardless of the tool used."
Despite these caveats, Subrt expects to see banks' adoption of these kinds of AI agents accelerate over the next 12 to 24 months, especially in enhanced due diligence and sanctions alert review, where the manual research burden can be the heaviest.
The banks that invest now "are establishing the operational standard against which others will be assessed," he said.
