Introduction

Amazon recently launched a feature where hundreds, and sometimes thousands, of reviews are summarized into a concise and simple paragraph.  This auto-generated summary allows the customer to instantly get an overall review of a product’s capabilities as well as overall customer satisfaction. This is a great example of using Large Language Models (LLM) to process and produce text that resembles that of humans. These models can understand language structures, grammar, context, and semantic linkages since they have been trained on enormous amounts of text data.

When it comes to fighting against financial crimes,let’s look at the use cases where financial institutions (FI’s) file Suspicious Activity Reports (SARs) to FinCEN. In the recent past, FinCEN has fined several FI’s to the range of millions of dollars for not filing SARs on timely basis. This has led to an increase in FI’s filing hundreds and thousands of SARs at an alarming rate.This problem has been compounded because firms use traditional rule-based scenarios to detect potential money laundering activity. The total number of SARs filed in 2022 surpassed 3.6 million, an increase of 57% from pre-pandemic, 2019 levels. FI’s usually need to have large investigation teams to review and address the alerts being generated by various transaction monitoring solutions.

Now imagine a future state, where generative AI based tools are assisting investigators by collating and summarizing data related to alerts and suspicious activity, allowing investigators to receive precise information and summaries, and helping draft narratives for regulatory reporting. This is already happening, and this phenomenon will pick up pace as many firms compete to adopt the latest technologies (including generative AI) in their solution offerings. With the rapid evolution of technology, there has been an increasing interest in how generative artificial intelligence (AI) models may be useful in combating financial crime.

US regulators are strongly encouraging FIs to adopt innovative solutions in combating issues related to Financial Crimes. As per 2022 National Illicit finance Strategy from Treasury Department, innovations in digital identity and AI, including innovative transaction monitoring systems and suspicious activity identification and reporting tools, can strengthen AML/CFT compliance and help banks and other FIs more effectively and efficiently identify and report illicit financial activity. In this article, we explore the benefits and limitations of using generative AI in detecting and preventing financial crime.

Generative AI Overview

Generative AI is a subset of artificial intelligence that uses deep learning algorithms to create new and unique content. Generative AI models are trained on existing content (data) and leverage statistical techniques to analyze patterns and relationships within the data. Once a model is trained, it can then be utilized to generate new content based on the patterns detected in the original, existing, dataset.

Generative AI for Combating Financial Crimes

Generative Adversarial Networks (GANs), Variational Autoencoders (VAEs) and Large Language Models (LLMs) are categories of generative AI being used in combating financial crime. 

• GANs are deep learning models that consist of two neural networks, a generator and a discriminator, which work in opposition of one another. The generator creates fake data, and the discriminator tries to distinguish between the real and fake data. The two networks are trained together with a goal of training the generator to create fake data that is indistinguishable from real data. 

• VAEs are another type of generative AI model that can encode and decode data from a given dataset. The model works to train the encoder side to compress given datasets and then trains the decoder side to reconstruct and generate new samples from the original, compressed, data.  VAEs can be trained on normal transaction data to learn the underlying distribution of legitimate transactions. Any deviation from this learned distribution can be flagged as a potential anomaly, which might signal fraud.
  This could be particularly useful for detecting new, previously unseen types of fraud. Anomaly detection is just one example of how VAEs may be used to assist in combating financial crimes.

• Large language models (LLMs) can be used to detect financial crimes, such as fraud, money laundering, and other transactional anomalies. LLMs can analyze large amounts of data and identify patterns that may be invisible to the human eye. By training AI models on large datasets of financial transactions, patterns of fraudulent activities can be identified and flagged for further investigation. They can be used in combination with other technologies, such as rule-based systems, supervised machine learning, and other specialized models.

GANs and VAEs can be used to generate synthetic financial data that can be used to train machine learning models. This synthetic data can be generated to simulate realistic fraudulent activities and train machine learning models to identify fraudulent patterns accurately. For example, let’s say a bank’s fraud detection system previously identified transactions over $1,000 as potentially fraudulent.

With the help of GANs, the bank can now train its system to identify new types of fraud, such as transactions that occur at unusual times of day or involve specific merchants or categories. Even if the bank hasn’t encountered any fraudulent transactions fitting these criteria, GANs can generate synthetic data that the fraud detection system can use to learn and detect these new types of fraud.

Utilizing generative AI can increase the accuracy of financial crime detection and prevent fraud from occurring in the future. GANs and VAEs can also be used to develop predictive models that can forecast future financial crimes. By analyzing patterns in historical data, AI can identify potential future threats, enabling investigators and FIs to take preventive measures.

Benefits of Using Generative AI for Combating Financial Crimes

There are several benefits of using generative AI for combating financial crimes, including:

• Improved Accuracy: By analyzing large amounts of data, generative AI can identify subtle patterns that indicate potentially fraudulent activity. In doing so, these models can increase the accuracy of fraud detection by identifying patterns and anomalies.
• Increased Detection Rates: Generative AI can identify patterns and anomalies in financial data that may be missed by traditional methods of detection, increasing the detection rates of financial crimes.
• Reduced False Positives: Traditional fraud detection methods can result in a large number of false positives, leading to unnecessary investigations. Generative AI can reduce the number of false positives by accurately identifying fraudulent activities.
• Faster Investigations: Generative AI can analyze vast amounts of data in real-time, enabling investigators to act quickly and prevent financial crimes from taking place.
• Predictive Capabilities: Generative AI can develop predictive models that can forecast future financial crimes, enabling investigators to take preventive measures.
• Improved Investigations: Generative AI can assist investigators in their work by providing valuable insights and analysis. By analyzing large datasets of financial transactions, generative AI can identify patterns and anomalies that may be missed by human investigators. This can significantly reduce the amount of time and resources required for investigations. Generative AI can also be used to automate certain aspects of investigations, such as data collection and analysis. This can free up resources to focus on higher-level tasks, such as developing strategies for preventing future financial crimes.
• Increased Accuracy and Consistency: Generative AI can improve the accuracy and consistency of investigations. Unlike human investigators, AI does not suffer from biases or fatigue and can analyze vast amounts of data with a high degree of accuracy.

• Increased Comprehensive Detection:  Generative AI can generate data and be trained to detect a wide range of financial crimes, including some behaviors/crimes that a bank has not encountered and therefore does not have the historical data to enhance threat detection capabilities.
• Efficient Scalability: Generative AI models are ideal for large FIs because they can easily scale up to handle large datasets.  

Limitations of Generative AI

Despite the potential benefits of generative AI in combating financial crimes, there are some limitations to consider, including:

• Data Quality Dependency: The effectiveness of these models is dependent on the quality and quantity of the data used to train them. Incomplete, biased, or inaccurate data may produce unreliable results.
• Computational Complexity: Generative AI models can require significant processing power, which can create difficulties when implementing on certain platforms or in certain environments.
• Ethical Considerations: Some ethical considerations raised include avoiding bias or discrimination and ensuring data privacy.
• Interpretability: Generative AI models can be complex, which may pose a challenge for FIs to understand and effectively use. 

Conclusion

Generative AI has enormous potential to revolutionize the way FIs combat financial crimes. By analyzing vast amounts of financial data and identifying patterns of fraudulent activities, generative AI can increase the detection rates of suspicious activities, reduce false positives, enable faster investigations, and provide valuable insights and analysis to investigators.

The predictive capabilities of generative AI can also be used to forecast future financial crimes, enabling FIs to take a proactive approach by designing preventive measures and implementing detection enhancements. As technology evolves, generative AI can be an indispensable tool for combating financial crime.

Authored by Omesh Bhatt, Managing Director, and Alexandra Methot, Senior Consultant, @ Matrix-IFS