Machine Learning and AI in Fraud

With so many advances in technology in the last few years, like Machine Learning and AI,  it is no wonder that criminals are able to harness it to commit wide scale fraud. However, the same weapons that scammers use to commit fraud can also be used to prevent and detect fraud. Here is what you need to know about machine learning and AI and their relationship to fraud and how you can use these tools to safeguard your interests.

Ways Machine Learning and AI Are Used to Commit Fraud

Technology has certainly made it easier for criminals to commit crimes. In 2015, 38% of consumer complaints were associated with fraud while 52% of them were associated with fraud in 2019, according to statistics of identity theft from the Insurance Information Institute.

Just a few specific examples of how machine learning and AI are used to commit fraud include:

• Forged signatures – A signature is any symbol that shows a person’s intent to be bound by the terms of a contract. If a signature can be successfully forged, the consumer can potentially sign any number of contracts. Researchers developed a program that was capable of successfully forging a signature when they had only one paragraph of a person’s writing.
• Impersonation of consumers – In “The Malicious Use of Artificial Intelligence: Forecasting, Prevention, and Mitigation” report, researchers warn that “the ability to generate synthetic images, text, and audio could be used to impersonate others online.” For example, fraudsters can use AI to commit a brute force attack, running through account numbers until they eventually get one. This happened to one company called Authorize.net in which fraudsters first tried to charge 1 cent to accounts and then later charged $7,000 to one merchant.
• Manipulation of AI to benefit fraudsters – AI uses complex algorithms to identify patterns and complete processes. Certified public accountants fear that the manipulation of AI can lead to fraud being perpetrated on consumers.
• Advent of cryptocurrency – New forms of cryptocurrencies continue to be developed, which may make them reachable and manipulated by hackers. For example, in one attack the same unit of currency was used multiple times, which resulted in more than $1 million in losses.

As technological advances continue to grow in number, so do the potential uses to use them to cause harm.

Internet of Things and Fraud 

With everything from faucets, toasters, security cameras, medical equipment, refrigerators, baby monitors, and watches being connected through the internet, there are more possible entry points for fraudsters, especially when these points are unsecured. Cybercriminals are able to commit fraud in novel ways, including the following real-life examples:

• A smart deadbolt had security issues that could allow attackers to remotely unlock doors, making the homes vulnerable to burglary.
• Cameras in hotels and Airbnb properties spurred privacy concerns because they could be hacked into. Hundreds of motel guests were recorded and live-streamed.
• A 14-year-old boy was able to use default IoT device settings to shut down 4,000 devices with a new strain of malware.
• A series of children’s smartwatches were found to be particularly vulnerable to hacker attacks, putting the children at potential risk by their GPS-enabled devices, spurring recalls.

While the IoT makes it easier to be connected, it can also impose vulnerabilities that subject consumers to the potential loss of their personal information, interference with the operation of their devices, and  danger to their homes and families.

Ways Machine Learning and AI Are Used to Prevent and Detect Fraud

On the other hand, machine learning and artificial intelligence are increasingly being used to prevent and detect fraud through methods such as:

• Using Machine Learning to Train Models

One way to combat the increased number of fraud attempts is to use machine learning to train models so that fraud can be more easily detected. This method is particularly effective when supervised. A computerized search for patterns occurs much more quickly than one from a human eye.

• Use Patterns Identification

Patterns identification can help prevent identity theft. Unsupervised machine learning can compare new consumer behavior to traditional trends to potentially detect fraud.

• Synthesize Supervised and Unsupervised Machine Learning

Many of the most effective fraud detection applications synthesize supervised and unsupervised machine learning. Supervised machine learning is trained on specified transactions, either identifying them as “fraud” or “non-fraud” transactions. Unsupervised machine learning is designed to identify behaviors that are anomalies based on historical data. This type of model relies more heavily on self-learning. By synthesizing the two models, programmers are often able to identify more problematic behaviors and spot fraud on a larger scale.

• Use ID Document Forgery Detection Models

This detection method analyses a consumer’s signature and looks beyond the top layer of information. Instead, the program checks for multiple layers and filters to help determine the presence of a forgery. This type of system can help determine a mismatch in ink and determine if the pixels in a photographed signature are not authentic.

• Strengthen E-Commerce Channels

With the advent of EFT technology, many attempts at fraud through the use of a physical credit card are down. However, at least 60% of credit card fraud cases in Europe derive from card not present transactions. This makes e-commerce platforms particularly vulnerable to this type of attack. Machine learning algorithms tend to detect fraud much faster without interfering with the completion of legitimate transactions.

• Develop Risk Profiles

Another machine learning based fraud prevention method is to develop risk profiles for potential targeting. This method is most often used in the banking industry in which banks use granular data to give customers a score. Their systems use machine learning that consider widespread fraud schemes.

These applications and others can potentially identify and prevent fraud.          

Conclusion

As long as there is a potential financial gain, fraud cases will continue. However, with an understanding of what types of fraud can be committed through the use of machine learning and AI and embracing effective technology that can prevent and detect it, you can place yourself ahead of the curve.

About the Author

David Lukić is an information privacy, security and compliance consultant at IDstrong.com. The passion to make cyber security accessible and interesting has led David to share all the knowledge he has.