Why We Need to Rethink AI in AML

By Anders Meinert Jørgensen - CEO and Co-Founder at Avallone

Artificial intelligence has been a recurring topic in anti-money laundering (AML) for years. For a long time, I shared the skepticism that many in the industry still hold today. That skepticism did not come from a lack of interest in innovation. It came from experience.

During my years leading compliance and financial crime prevention functions in large financial institutions, including as Chief Compliance Officer at Danske Bank, I saw firsthand the practical limitations of early machine learning models in AML.

Two fundamental challenges with AI in AML

Explainability

The first challenge was explainability. The models we evaluated could generate outputs - risk scores, alerts, classifications - but they could not clearly explain how those outputs were reached. In a regulated environment, that is not a technical inconvenience. It is a critical failure point. Every decision in AML must be documented, defensible, and auditable. Regulators expect clear reasoning, not black-box conclusions.

Data Quality

The second challenge was data quality. AML and KYC data is rarely clean or consistent. It is shaped by years of evolving regulation, shifting internal policies, and different interpretations across teams and jurisdictions. The same activity may have been assessed differently over time. Training models on that historical data effectively means training them on inconsistency.

“At the time, challenges with explainability and data quality were not theoretical concerns. They were operational realities. And in that context, skepticism toward AI was justified.”

What has changed is not the regulatory environment, but the technology

The emergence of large language models (LLMs) represents a meaningful shift in how AI can be applied in compliance. Unlike earlier machine learning approaches, LLMs are capable of structuring explanations, referencing the data they draw upon, and making their reasoning visible to the user.

In AML, auditability is non-negotiable. Analysts must be able to understand, challenge, and document the reasoning behind decisions. LLMs do not remove that requirement, but they do offer a way to support it. Instead of acting as opaque decision-makers, they can function as tools that enhance human judgment while maintaining transparency.

Data quality challenges remain. Most financial institutions still operate with fragmented systems, duplicated records, and historical inconsistencies. However, LLMs appear better equipped to handle imperfect data. They can interpret context, work with unstructured information, and identify inconsistencies across sources. There is also a growing case that they can help improve data quality over time by standardizing how information is processed and interpreted.

The more important shift, however, is strategic.

For years, the AML industry has taken a cautious approach to AI. That caution was understandable. The risks were real, and the technology was not mature enough to meet regulatory expectations. But today, caution is no longer a neutral position.

The volume of data is increasing, financial crime is becoming more complex, and regulatory expectations continue to evolve. At the same time, compliance teams are under pressure to deliver more with limited resources. In this environment, relying solely on manual processes and fragmented systems is not sustainable.

Choosing not to adopt new technology now carries its own risks. Inefficiencies, inconsistencies, and lack of visibility can lead to operational failures and regulatory exposure.

The question is no longer whether AI should be used in AML, but how it should be implemented responsibly

A needed shift in mindset for AI and AML

AI should not be viewed as a replacement for compliance expertise. It should be seen as an enabler. The objective is not to automate decisions blindly, but to provide analysts with better tools, clearer insights, and more structured workflows.

This also means being pragmatic about where AI can add value. In my view, the most immediate opportunities lie in areas such as document analysis, customer due diligence support, and alert investigation workflows. These are processes that are currently resource-intensive and often inconsistent, where improved structure and efficiency can have a meaningful impact.

In our work at Avallone, we see this particularly in how organizations manage KYC data and documentation. A simple example is handling and responding to bank KYC questionnaires, where teams often spend hours searching across emails, shared drives, and internal systems to find the right information. Structured workflows combined with AI-supported analysis can significantly reduce that effort by extracting, organizing, and surfacing relevant answers - while still allowing the user to verify the source before responding.

The most effective use cases are those where AI acts as a “wingperson” to the analyst - extracting and structuring data, surfacing relevant information, and suggesting responses, while keeping the human fully in control.

Equally important is explainability. For AI to be viable in compliance, its outputs must be transparent and traceable back to source data. The objective is not just automation, but to build systems where users can verify, challenge, and trust the results - even when working with imperfect data.

At the same time, governance remains essential. Transparency, auditability, and clear documentation must be built into any AI-supported process. The regulatory expectations have not changed, and they should not.

Looking back, I believe the skepticism many of us had five years ago was justified. Concerns around explainability and data quality were real. What’s changed is that the technology has evolved - and we as an industry must evolve with it. The conversation should no longer focus on whether AI can be used, but on how it can be used effectively, responsibly, and in a way that strengthens the overall compliance framework.

Anders Meinert Jørgensen, CEO, Avallone

A successful corporate banker for +20 years, Anders Meinert Jørgensen fought financial crime first-hand while leading compliance divisions in two of the largest banks in the Nordics. Realizing that the industry is hindered by a lack of efficient tools, Anders co-founded Avallone to give companies and financial institutions a world-class platform that empowers them to handle KYC and combat financial crime.