According to a new study released by the Association of Accredited Fraud Examiners (AAFE) , anti-fraud experts are seeing a spike in different forms of fraud due to the COVID-19 pandemic. As of November 2020, the survey reports that 79% of respondents said they have seen an increase in the overall level of fraud, with 38% noting that this increase has been significant (compared to 77% in August and 25% in May). As we look toward 2021, the AAFE team expects this trend to persist; 90% anticipate a further increase in the overall level of fraud over the next 12 months, with 44% saying this change will likely be significant.
What kinds of frauds are on the rise?
Cyberfraud (e.g., business call/email compromise, ransomware, and malware) continues to be the most heightened area of risk, with 85% of respondents already seeing an increase in these schemes, and 88% expecting a further increase over the next year. Other significant fraud risks in terms of both observed and expected increase include payment fraud (e.g., credit card fraud and fraudulent mobile payments), identity theft, and financial reporting fraud.
Considering the rise in the overall risk of fraud, and changes in particular fraud risk categories, organizations need to ensure that their anti-fraud programs are more real-time and effective. New threats are difficult to detect when an organization depends on rule engines or experts alone. AI will be the perfect match for battling challenges posed by fraudsters in these challenging times. AI can beat the traditional approaches and mitigate risk with quicker turnarounds. Here are a few areas where AI could help battle frauds.
How AI can help battle
Tax evasion refers to the deliberate act of lying on a tax return form to minimize one’s tax liability. In the UK, HMRC confirmed many fraudulent claims related to the furlough scheme (Coronavirus Job Retention Scheme) in which it had received 795 reports of potential fraud. It also reports that, as of 29 May 2020, it had received 1,868 such reports. Misuse of the scheme will likely involve tax fraud as the scheme gives exemption for National Insurance Contributions (and accordingly tax will have been evaded) .
With an AI solution that combines data sources from different owners such as real estate, utilities, banks, customs etc., and processing unstructured data such as descriptions of tax policies, payment documents using NLP and machine learning techniques, we can identify the likelihood of delinquency and fraud for deeper analysis.
During the pandemic, organizations all over the world have moved towards virtual onboarding. This might have different phases such as coding exams, interviews and require authentication of the users. The ability to confirm the identity of a user with facial recognition may appear to be the effective authentication method in today’s world.
In most cases, image manipulation techniques can be used by fraudsters to create false IDs, using which they can illegally gain access to a device or service. Liveness detection techniques can be adopted to keep such fraudulent activities in check. These techniques can determine if it is a physically present human being or an inanimate spoof item, which can help curb fraudsters from accessing or constructing online accounts using stolen images, deep fake videos, or masks. This technology makes use of computer vision algorithms to determine if the image is flat, analyze facial movements (including head movement, eye blinking, and mouth opening) to determine liveness.
Recently  fraudsters have used robocalls to impersonate government investigators and to scare Americans with alarming messages. Calls facilitated by the defendants threatened victims with a variety of catastrophic government actions, including termination of social security benefits.
The typical process is to ask questions to confirm the identity of the person at the beginning of the call. Historical analytics on these fraud calls can result in understanding the questions often posed by fraudsters. Too often these questions may be “Can you confirm your phone number?” or “What is your address?”. The issue with these types of questions is that this information is readily accessible online and fraudsters easily gather it before they make a call to gain the trust.
A caller is classified as a fraudster on the basis of his features derived from the CDRs (customer details records) which represent the logs of the following event within a specific timestamp. Speech analytics solutions can review the data and trends to determine which keywords and phrases you should create notifications for. Other signs that the solution can consider include whether the caller has contacted many numbers more than once within a certain number of days and the number of international calls, and an alert can be created accordingly.
Latest reports show a 273% rise in large-scale data breaches in the first quarter of 2020, relative to the previous year’s figures, and a 109% year-on-year increase in ransomware attacks in the United States in the first half of 2020 . Hackers use phishing or other means to place malware on the victim’s computer system. This encrypts the system, making the system files and data inaccessible to the victim. The hackers then attempt to obtain monetary payment from the victim in return for the key needed to decrypt the compromised data.
Machine learning models that analyze URL prefixes, suffixes, length, shortening services, valid domain registry can be a huge value-add to emails about the possibility of phishing.
With economies all over the world taking a hit and many companies already experiencing a significant decline in revenues, loss from cyberattacks at such times can be very expensive. Therefore, there is an urgent need for companies to protect data, monitor for fraudulent activities more rapidly, utilize AI solutions that expose early potential frauds and help prevent them.