Saturday, July 20, 2024
- Advertisement -

    Latest Posts

    Here’s how AI is tackling money laundering through cryptocurrency

    The rise in cryptocurrency’s popularity has led to a new form of cybercrime. The partial anonymity offered by cryptocurrency has lent itself to perpetrators of financial crimes such as money laundering. Law enforcement is often at a disadvantage when it comes to identifying cases of money laundering through cryptocurrency, as it is extremely challenging to trace suspects from a large amount of data on the blockchain.

    However, AI and machine learning are particularly equipped to analyse large amounts of data. Thus, there have been notable developments in the field of AI to tackle the problem of financial fraud through cryptocurrency. Notably, Elliptic, a cryptocurrency intelligence company focused on safeguarding cryptocurrency ecosystems from criminal activity released a paper along with MIT-IBM Watson AI Lab. The paper explores the workings of a machine learning model that can identify transactions that can be instances of money laundering. The data set could identify a flow of Bitcoins that may be linked to money laundering activity, by detecting instances of a cryptocurrency chain being converted into legitimate currency. This data set, called Elliptic2, was made available to the public in order to encourage further research into financial crime detection tools.

    Some context on money laundering

    There are multiple methods of laundering money on the blockchain. Bad actors could use cryptocurrency to convert their laundered money into cash and legitimise the transaction. They can also transfer funds overseas and to other accounts without being traced. Often, fraudsters create wallets to make multiple transactions of small amounts from their large sums of money. This makes it exponentially harder for law enforcement to trace back to a single criminal. According to Chainalysis, in 2023, more $22.2 Billion worth of cryptocurrency was sent from illicit services, indicating money laundering activity.

    There have been regulations set into place to tackle this. The reason Bitcoin is not anonymous but rather pseudo-annonymous is because in the United States, according to the guidance issued by the Financial Crimes Enforcement Network (FinCEN) on how the Bank Secrecy Act (BSA) of 1970 applies to cryptocurrency, networks are required to “know enough about their customers to be able to determine the risk level they represent to the institution.”

    Networks are required to have an anti-money laundering (AML) program as per FinCEN. This requires them to conduct an adequate risk assessment of their network and determine both the identity and profile of its customers. This can often require them to conduct Know your Customer (KYC) services and classify accounts that may be illicit. However, there are significant challenges in this process. Particularly, with contacting individuals. Likewise, identifying illicit transactions from large ever-growing datasets. This can often lead to high false positive rates that can be a burden on time and resources. Thus, AI can be used to decrease the labour associated with this process.

    Classifying illicit transactions using AI

     In 2019, Elliptic published a paper detailing how machine learning could be used to classify accounts as illicit and licit, based on the history of their transactions. Elliptic used the large amounts of Bitcoins’ publicly available raw transaction data to create a data set. The data set consisted of 200 K Bitcoin transactions over a specified period of time. The data set was “a graph network” of Bitcoin transactions which were labelled to be either “licit’ or “illicit”. Transactions that were linked to entities such as exchanges, wallet providers, miners, and other reputable sources were classified as licit. Similarly, transactions associated with scams, malware, terrorist organizations, ransomware, Ponzi schemes, etc. were tagged as illicit.

    The set created these binary classifications through a “heuristics based reasoning process”. For example, accounts which reuse the same address and have a higher number of inputs are commonly associated with legitimate activity. This is because these transactions reduce anonymity for the entity signing the transaction. Similarly, an account that consolidates funds controlled by multiple addresses in one single transaction and in turn reduces anonymity-preserving measures for large volumes of users are likely to be legitimate exchanges.

    Conversely, accounts that tend to favor transactions with a lower number of inputs reduce the impact of de-anonymizing. Thus, they are more likely to be illicit. Elliptic used Logistic Regression (LR), Random Forest (RF),  Multilayer Perceptrons (MLP), as classification techniques and employed Graph Convolutional Networks (GCN) for scalability. The research noted that Random Forest significantly outperformed Logistic Regression and GCN.

    At the time, this dataset was the largest labeled dataset of Bitcoin transactions. It was subsequently made publicly available to encourage further research. It enabled tracing accounts that were associated with illicit activity.

    Identifying money laundering transactions

    On May 1, 2024 they published another paper on enhancing money laundering detection using AI. The new dataset combats financial crime by identifying whether a specific flow of bitcoins may be linked to money laundering activity rather than identifying transactions made by illicit actors. They do this by observing if the transactions follow odd patterns and if cryptocurrencies are converted into cash after multiple points of transaction.

    The model uses “Subgraph representation” which is a technique for analyzing local structures (or shapes) within complex networks. This technique can identify an unusual series of chain transactions or “shapes” that distinctly resemble money laundering patterns.

    The paper is based on the theory that “a path on the blockchain connecting an illicit cluster to a licit cluster without a change of ownership of the funds likely represents the activity of money laundering by a criminal person or organization.” Thus, the chain of transactions from licit accounts to illicit accounts, called a “multi-hop” laundering process, can create a “shape” that can be identified as subgraphs known to be linked to illicit activity like money-laundering.

    Elliptic2 is a dataset of 200-million classified and labelled Bitcoin transactions. The dataset defines an illegitimate transaction by the time window, the maximum number of hops between accounts, and the conditions when a change of ownership is likely to happen. For this dataset, the time window was chosen to be 1 year of blockchain data, and the maximum number of hops was 6.  It also called to define each step of the transaction defining it as “licit” or “illicit.”

    3 methods were used to train the model- GNN-Seg, Sub2Vec, and GLASS. All methods were able to converge in several days of training with an inference time that was less than 8 hours. However, GLASS was used for further experimentation.

    For the experiment, 52 subgraphs that were deemed suspicious and which were cashed out at an exchange were chosen. The exchange was then asked to conduct an assessment of these accounts’ legitimacy based on their due diligence practices. According to the exchange,14 of the 52 accounts were potentially involved in illicit activity.Similarly, when the model was tested at scale for another experiment, investigations were undertaken to identify the origin of funds flowing into subgraphs deemed suspicious. Of these subgraphs, 182 were identified to be associated with financial fraud.

    Additionally, the model was also able to identify known methods of money laundering- peeling chains and nesting services. “Peeling chains” entails a user reducing or “peeling” a small amount from a large amount to a separate address, while the remainder is sent to an address belonging to the user. This chain continues, with the amount becoming smaller while making the chain getting harder to trace. Eventually, fraudsters cash these amounts at exchanges. “Nested services” on the other hand are businesses that hold accounts at larger cryptocurrency exchanges and enable liquidity for accounts without direct interaction with the exchange. These can be abused by money launderers to cash out their cryptocurrency without directly transacting with the exchange.

    The Elliptic2 data set and model has been made widely available to encourage further research.

    Why it matters

    Money laundering through cryptocurrency has been a major concern for law enforcement and there has been an increasing demand for technology to stay abreast. For example, recently Chennai Police released a tender calling for a tool to analyse cryptocurrency transactions to tackle financial fraud.

    Further, while preventative measures like KYC may be necessary to identify illicit activity, it defeats the purpose of anonymity promised by cryptocurrency. Thus, developments in AI in this field are significant.

    Also Read:

    STAY ON TOP OF TECH NEWS: Our daily newsletter with the top story of the day from MediaNama, delivered to your inbox before 9 AM. Click here to sign up today!


    The post Here’s how AI is tackling money laundering through cryptocurrency appeared first on MediaNama.

    Latest Posts

    - Advertisement -

    Don't Miss

    Stay in touch

    To be updated with all the latest news, offers and special announcements.