How Generative AI Enhances AML, Sanctions Screening, & Fraud Prevention

Generative AI has the power to transform Anti-Money Laundering (AML) Transaction Monitoring, Sanctions Screening, and Fraud Prevention. In fact, we predict that Generative AI will soon be essential to AML, Sanctions, and Fraud teams aiming to detect and manage emerging risks.

Globally, financial institutions (FIs) are using AI to maintain regulatory compliance, prevent and detect financial crime, and manage false positives. The impact has been significant – Hawk customers are seeing 3-5x increase in risk detection and 70% reduction in false positives. Generative AI shows great potential to deliver even greater benefits.  

Key Takeaways

1. Generative AI can have an even greater impact than classic machine learning on the efficiency and effectiveness of AFC operations
2. FIs can use generative AI for:
   1. Sanctions Screening
   2. Fraud Prevention and AML Transaction Monitoring
   3. Case Investigation
   4. QA Processes

What Is Generative AI, and How Is It Different from Other AI Technologies?

Generative AI is an umbrella term for AI models that can generate content. The most prominent examples of Generative AI are Large Language Models (LLMs). These models can interactively create text and answer questions. Other common Generative AI use cases include generating images, videos, and music. Generative AI technology relies on advanced machine learning techniques, including Transformer Neural Networks, Generative Adversarial Networks (GANs), and Variational Autoencoders (VAEs).

These models can process and generate content in response to user prompts or predefined criteria. The models understand the structure of language, generate coherent and relevant text, and even answer questions. With the massive volumes of text data available on the internet, Generative AI has seen major breakthroughs in the last few years. 

Although there are overlaps in the underlying AI models applied, Generative AI differs quite profoundly from classic machine learning technology—from its objectives to its applications. With regards to objectives, Generative AI is designed for (but not limited to) creating new content, such as text, images, or music. In contrast, classic machine learning is designed for making predictions and classifications based on existing data. In terms of applications, we are now seeing promising results using Generative AI for predictive purposes, specifically in the realms of Sanctions, AML, and Fraud risk.

What Impact Will Generative AI Have on AML, Sanctions Screening, and Fraud?

Classic AI is already helping AFC teams to improve results, particularly in false positive reduction and anomaly detection. We have seen successful implementations along a sophistication path that is driven by trust in AI:

There are four areas in which AI will continue to have a significant impact on AFC, with additional lift from Generative AI:

1. Detecting Sanctions Violations 
2. Detecting Money Laundering and Fraud Activity
3. Aiding Case Investigations
4. Improving QA Processes

1. How Does Generative AI Improve Sanctions Screening?

Scanning and interpreting text related to transactions or trades is essential to detecting sanctions evasion. Investigators need to scan this data for links to sanctioned entities or dual-use goods. 

Looking for risky words, dual-use goods terms, or other clues in messages inherently creates huge amounts of false positives. A human investigator can easily determine that these alerts don't point to suspicious activity, but human teams do not have the time to work through hundreds of false positive cases. Classic machine learning algorithms don’t understand the semantics of language and just try to fuzzy match, which doesn't move the needle much. That's where Generative AI comes in.

Generative AI excels at working with unstructured data like this. Generative AI can discard alphabetically similar, but semantically different, false positives while finding synonyms or slang terms that are not explicitly listed as of concern. If “drug” is on a risky words list, the LLM would identify “hemp” or “snow” as synonyms, even though they are not on the list. This capability will drastically improve the effectiveness of screening, as well as making managing screening lists simpler and less time consuming.

2. How Does Generative AI Improve Fraud Prevention and AML Transaction Monitoring?

Classic AI models have improved AML and Fraud detection on many fronts. From alert prioritization and filtering, to pattern and anomaly detection, to AI-only solutions such as the HSBC Transaction Monitoring deployment, AI has boosted the effectiveness and efficiency of AML and Fraud operations. At Hawk, we have seen false positive reduction rates of more than 70%, as well as four times as much suspicion on AI alerts over rule-based triggers.

Generative AI can enhance these use cases significantly, both increasing detection rates and reducing false positive reduction rates. We can train Generative AI models on large volumes of transaction data, teaching them the “language and grammar” of transactions. The model develops a deeper understanding of customer behavior by analyzing transactions and their correlations, just as an LLM like ChatGPT develops a deeper understanding of natural language. It can identify complex relationships between transactional attributes better than classic machine learning technology can. These analytical capabilities enable Generative AI to understand AML and Fraud risk at previously unseen depths. With Generative AI, human operators can manage risk more effectively and efficiently than ever. 

“A Generative AI model’s in-depth representation of transactions allows it to see links in transactional activities,” said Wolfgang Berner, Co-Founder and CPO at Hawk. “For example, it sees the relationship between money moving into an account and then moving out quickly. It understands that, in financial crime risk terms, this can mean potential money mule behavior. It will even see the correlations if the incoming or outgoing money is split up, looking at timing, sequence, and other hidden clues in the data.”

3. How Does Generative AI Improve Case Investigation?

Since ChatGPT has reached widespread adoption, we have seen the power of Generative AI creating high quality textual content and refining such content interactively through prompts and speech. Generative AI-based co-pilots have proliferated, helping users to complete tasks more quickly and effectively with their deep understanding of language. 

In a financial crime case investigation context, a co-pilot can help create case narratives. When we feed background and investigation history into a Generative AI model, we can achieve effective results. Human oversight on such narratives and summaries are key. It is important for the human users to understand the underlying facts contributing to the summary. The human investigators may even need to validate the facts in detail before submitting to the Financial Investigation Unit (FIU).

We also can see co-pilot type applications that allow an investigator to dig deeper on certain risk. Generative AI co-pilots can also source additional contextual information from the web, social media, and map search, as well as from other areas of the business, such as relationship management. These co-pilots can formulate questions, analyze, and summarize outcomes, helping to facilitate and streamline investigative processes.    

4. How Can Generative AI Improve QA Processes?

Larger, geographically distributed investigation teams often struggle with stable quality and consistency in case investigation. Depending on training, background, and experience, their investigation results can vary. We can apply Generative AI to enhance QA processes in the following ways: 

• Language processing: Generative AI can analyze case outcomes (compared to triggers) and the narratives created across teams. This helps FIs determine deviations that can be addressed downstream with additional training measures for the teams. 
• Identifying Detection Deficiencies: Generative AI can find differences in investigation tooling or processes. This helps FIs mitigate any deficiencies identified in their AFC programs.
• Internal Reporting:  Generative AI can create summary reports on QA and other general reports for management. This will help AFC teams increase visibility and transparency with the FI’s internal stakeholders.

All these use cases will serve to improve the quality of Sanctions, AML, and Fraud operations while streamlining AFC processes. 

Hawk’s AI-Powered Sanctions, AML, & Fraud Risk Detection Technology

Hawk’s AI-powered Sanctions, AML, and Fraud risk detection technology helps FIs improve the efficiency and effectiveness of their AFC operations. FIs can leverage Hawk technology’s false positive reduction and anomaly detection capabilities to improve their risk coverage while using fewer resources. We build data privacy controls and explainability into our technology to help FIs manage risk as securely as possible.

To learn more about Hawk’s AI Sanctions, AML, and Fraud risk detection technology, request a demo today.