AML compliance: money laundering prevention with AI

With the gradual implementation of the 6th EU Anti-Money Laundering Directive (6th AMLD) and the stricter requirements of the Anti-Money Laundering Act (AMLA), financial institutions are facing the challenge of meeting increasingly complex AML compliance requirements in a cost-efficient manner. At the same time, modern AI technologies, particularly in the area of automated document analysis, are opening up new possibilities for risk assessment and auditing. This multi-layered interaction between advancing technology and European regulation has long since begun to spill over into the global financial sector, as several experts have pointed out.

The regulatory foundation: From the 5th to the 6th AMLD

The latest EU money laundering directives have significantly increased the requirements for due diligence. In particular, the provisions on beneficial owners under Section 10 (1) No. 2 of the German AMLA and Section 11 AMLA now require a continuous, risk-based review of complex ownership structures. The threshold for identifying beneficial owners of companies remains unchanged at 25% of the capital or voting rights, but the documentation and verification obligations have been significantly tightened.

At the same time, the increased due diligence obligations pursuant to Sections 15 et seq. GwG for politically exposed persons (PEPs) and business relationships with high-risk countries have reached a new level. Continuous monitoring in accordance with Section 10 (1) No. 5 GwG no longer requires only ad hoc, but permanent, risk-based controls of the customer base. This dynamic is also being observed outside Europe.

"This increase in regulation has caused financial institutions to become more proactive in their model of compliance. What was once accomplished through manually reviewing documents now involves perpetual monitoring of the transactional behavior, complicating it further and making it the responsibility of the perpetrator. The goal is not to determine obvious risks, but also to determine the subtle inconsistencies that might have been previously neglected" explains Paul Ferrara, Senior Wealth Counsellor at Avenue Investment Management.

Cost efficiency versus regulatory AML compliance

The fundamental issue of cost efficiency is also reflected in industry data: McKinsey determined in its 2024 KYC/AML benchmark studythat banks normally use 10 to 15 percent of their full-time equivalents for KYC/AML processes alone - an enormous cost factor that increases exponentially with manual processes. These personnel costs are further increased by the intensified documentation and monitoring obligations of the 6th AMLD.

Smaller institutions in particular are coming under cost pressure: while large banks can spread their compliance costs over a high transaction volume, the fixed regulatory requirements lead to high unit costs for smaller players. BaFin's Minimum Requirements for Risk Management (MaRisk) demand adequate human and technical resources. In practice, this means that institutions must prove that their compliance processes are state of the art - a criterion that is becoming increasingly difficult to meet without AI support. Technologies that specialize in the analysis and processing of complex documents are establishing themselves at the heart of the technical possibilities.

AI-supported document analysis

Document analysis forms the backbone of modern money laundering prevention and enables financial institutions to efficiently and precisely fulfill the extensive due diligence obligations under the AMLA. Two key technologies are used here: Optical Character Recognition (OCR) for text recognition and Natural Language Processing (NLP) for semantic understanding of the content.

The challenge lies in the diversity of the documents to be analyzed - from handwritten shareholder agreements to official register excerpts and international Identity documents. This requires highly developed AI systems that can reliably capture both structured and unstructured information.

Optical Character Recognition (OCR)

Modern OCR systems use convolutional neural networks and achieve a text recognition accuracy of over 99.5%. They form the first stage of an intelligent document processing workflow that is specifically designed to meet the requirements of money laundering prevention.

Core functions for the prevention of money laundering:

• Multi-format recognition - Automatic identification of ID, commercial register and articles of association documents in accordance with §§ 10-12 GwG
• Safety feature detection - Algorithms for recognizing watermarks, holograms and microprinting in official documents
• Multilingual processing - Support for over 100 languages, essential for international customers
• Quality assessment - Automatic check of document completeness for due diligence obligations
• Batch processing - Simultaneous processing of multiple documents for efficient onboarding handling

The precise text extraction enables fully automated documentation obligations in accordance with § 8 GwG. Each step is documented with a time stamp, which serves as proof of proper procedures during supervisory audits.

Natural Language Processing (NLP)

Modern NLP systems for detecting money laundering are based on Large Language Models, trained on financial and legal documents. They extract structured data from unstructured texts and automatically assign them to the corresponding compliance categories.

Extended NLP functions for AML compliance:

• Named Entity Recognition - Automatic identification of persons, companies and PEP-relevant titles
• Relationship extraction - Recognition of ownership and control relationships for determining the beneficial owner in accordance with Section 3 GwG
• Risk analysis - Evaluation of media reports on automatic risk classification
• Consistency check - Automatic comparison between different documents to identify contradictions
• Regulatory Mapping - Assignment of extracted information to specific MLA requirements

Practical application: NLP systems can automatically extract from a partnership agreement that "Mr. Max Mustermann with 30% participation" is the beneficial owner, transfer this information to the compliance system and perform PEP list comparisons at the same time.

The combination of OCR and NLP reduces manual document checking from hours to minutes, while increasing the quality of the compliance check - crucial for the increased requirements of the 6th AMLD.

KYC and identification of the beneficial owner

A typical use case: determining the beneficial owner in accordance with Sections 10 and 11 AMLA is possible in a multi-stage, AI-supported process:

1. automated data collection and completeness check

The AI system first checks whether all the required documents are available in accordance with the technical specification. The following sources are automatically queried:

• Commercial register (electronic register excerpt)
• Transparency register according to § 18 ff. GwG
• Company databases (Creditreform, Schufa company information)
• BaFin company database for licensed financial service providers

2. AI-supported structural analysis and investment tracking

Machine learning algorithms analyze complex investment structures and create them automatically:

• Graphical organizational charts with share distributions
• Shareholding chain analyses down to the natural person
• Risk scores based on structural complexity and geographical factors
• Cross-reference checks between different data sources

3. continuous monitoring and anomaly detection

The system monitors continuously:

• Changes to the commercial and transparency register
• Media screening for negative reports
• PEP lists and sanctions list updates
• Changes in corporate structure

Austin Rulfs, founder of Zanda Wealthreports from the field: "Our KYC and beneficial owner checks are done using document reading models, network matching, and selfie liveness. The shift reduced false positives by 41 percent to 17 percent, reduced average onboarding time by 48 hours to 6 hours, and reduced unit cost by 37 to 14 on 10,000 files."

Cross-border challenges

Correspondent banking is a particular focus of international money laundering prevention, as it is traditionally misused for complex, cross-border money laundering schemes. The FATF recommendations 13 and 16 require enhanced due diligence obligations for correspondent banking relationships, in particular the assessment of the correspondent's AML controls and continuous monitoring of business relationships with clients.

Matt Woodley, founder of InternationalMoneyTransfer.comexplains: "Stricter regulations and requirements are putting pressure on businesses to be even more vigilant in monitoring transactions and verifying customers. Given that cross-border transactions are at the core of the problem, regulations are a moving target."

AI systems can support this in particular by

• Real-time sanctions screening - Immediate comparison against all relevant sanctions lists (EU, OFAC, UN, national lists)
• Network Analysis - Detection of suspicious transaction patterns in correspondent banking networks
• Country Risk Assessment - Dynamic assessment of country risks based on current FATF assessments and political developments

Predictive analytics and advanced pattern recognition

In order to meet the increasingly proactive measures required to combat money laundering, analysis processes are used that assess transaction data and risks in real time. Modern AI systems use various machine learning techniques to ensure compliance:

• Unsupervised Learning - Clustering algorithms for detecting anomalous transaction patterns without prior labeling
• Supervised Learning - Random forest and gradient boosting for the classification of suspicious activities
• Deep Learning - Neural networks for the analysis of complex, multidimensional data structures
• Time Series Analysis - LSTM networks for the detection of temporal anomalies in transaction behavior

These technologies allow the creation of dynamic risk profiles based on:

• Historical transaction behavior
• Geographical risk factors
• Sector-related risk indicators
• Network effects (connections to other risk customers)
• External data sources (media reports, sanctions list updates)

Regulatory perspective and BaFin expectations

In its interpretative notes on the German Money Laundering Act (GwG) and in the banking supervisory requirements (e.g. MaRisk and BAIT), BaFin emphasizes that institutions must ensure a risk-based and effective use of IT systems. This includes in particular

• MaRisk amendment 2021 -Stronger requirements for outsourcing, IT risk management and data quality, which are also directly relevant for money laundering prevention systems.
• BAIT (banking supervisory requirements for IT) -Governance, data quality and integrity requirements for IT systems, which may also affect transaction monitoring and sanctions list screening.
• FIU requirements for suspicious activity reports (Section 45 GwG and GwGMeldV) -Technical standards for the electronic transmission of reports to the Financial Intelligence Unit.
  

The supervisory authority not only assesses formal AML compliance, but also the appropriateness, effectiveness and efficiency of the systems used. This increases the pressure on companies and institutions to further develop their processes technologically. Although BaFin does not explicitly prescribe the use of artificial intelligence, AI will be a decisive lever in practice in order to meet the increasing regulatory requirements for data quality, monitoring efficiency and risk orientation.

Austin Rulfs predicts: "The future is a split model. Direct through computer checks will clear approximately 90 percent of machine-assisted, clear, and clear-cut cases. Traditional verification remains in the final 10 percent where paperwork is incongruent, ownership is stacked among three or more players, or humanity requires someone to call the ball."

Conclusion

The increasing regulatory requirements, in particular due to the 6th EU Anti-Money Laundering Directive (6th AMLD) and the German Money Laundering Act (GwG), present financial institutions with the challenge of efficiently and precisely structuring their AML compliance. The use of modern technologies such as Artificial Intelligence (AI), Optical Character Recognition (OCR) and Natural Language Processing (NLP) offers a transformative solution.

By using such technologies, complex shareholding structures can be analyzed, beneficial owners identified and extensive documentation obligations under the AMLA fulfilled automatically. Even if the use of AI is not explicitly prescribed, it is already an economic necessity in many cases in order to remain cost-efficiently compliant.

Are you looking for technological possibilities for more efficient testing processes and AML compliance? Feel free to send us a Message, to obtain non-binding advice.