Fraud Trends 2023: What They Really Mean for Fraudsters and Your Business
Published on February 9, 2023 by Tamas Kadar
You would need to be hiding under a rock not to have heard industry insiders’ warnings that fraudsters are expected to benefit from the leaps in artificial intelligence this year. But, reading between the buzzword-filled lines, the real question is around how this will change the threat landscape.
Let’s explore this and other 2023 fraud trends in combination with internal SEON data, to see whether they add up and what they could really mean for organizations.
What Are the Most Prominent Fraud Trends for 2023?
Upcoming fraud trends revolve around increased sophistication on both the fraud-fighters’ and fraudsters’ side, as well as merchants’ increased focus on friction and how it is affecting fraud strategies:
- AI and Other Types of Sophisticated Fraud Are on the Rise
- There Are Lower Barriers to Entry and More Incentive for Fraud
- Concerns about Friction & Churn Will Provide Opportunities for Fraud
- Cross-Sector Collaboration Will Continue to be Key
- Anti-Fraud Machine Learning and AI Is about to Skyrocket
With AI and More Tools, Fraudsters Are Becoming More Sophisticated
Everyone’s talking about ChatGPT – from marketers to gamers and programmers. But, in fact, it’s not as new as one might think:
The machine learning technology behind it was initially released in 2020, with an API version in 2021 – while the current version became available on March 15, 2022. Dubbed GPT-3.5, it’s a language prediction and generation model that uses deep learning to understand and produce text that is, most of the time, convincingly human-like.
Moreover, image generation models like MidJourney are increasingly used in work and in leisure – and can also be used to generate deepfakes in the sense of mimicking a real person and placing them in specific situations.
Leverage unique digital footprinting and robust device fingerprinting in real-time to spot risky users and stay safe and compliant, without false positives.
Book a Demo
On January 2, Frank on Fraud’s Frank McKenna – an undeniable authority on the industry – released his annual fraud predictions. First and foremost were “shapeshifters”, a term to describe those who shift and change by the second thanks to AI and other tools, in order to better exploit gaps in defenses and differentiate their attacks.
But what can these machine learning modules actually do for a fraudster? This is the real question. Let’s take a closer look.
|More efficient automation of attacks||ML in combination with fraudulent bots and other automation can enable fraudsters of all persuasions to carry out more coordinated attacks in a way that appears to be more human and more genuine.||Better evasion of rudimentary velocity checks – with multi-accounting fraud on the rise as a result.|
|Diversified, bespoke phishing||Large language models can be utilized to quickly write and/or adapt social engineering attempts, including phishing emails and attempts at spear-phishing in a believable style.||Phishing and its sub-types of attacks are expected to become even more prevalent and believable – and as a result, so will identity theft, card fraud and everything they enable.|
|The creation of malware and other fraud-enabling tools||According to security company Checkpoint, bad actors have already started experimenting with the use of ChatGPT to code malware – specifically, with recreating existing malware strains and introducing new versions of them.||New malware, threatware, ransomware and other cybercrime tools will make their appearance – to help with phishing, take over accounts, block access to systems, sabotage companies’ infrastructure, etc.|
|Convincing fake documents created en masse||Image creation modules can be instructed to quickly and efficiently create believable identity documentation including driver’s licenses, scans of passports, ID cards, etc.||More prevalent synthetic identity fraud, with fraudsters more easily passing through simpler IDV checks to apply for financial products and grants, open bank accounts for money laundering, defraud iGaming companies, etc.|
|Improved deepfakes for image and video||Similarly, deepfakes are increasingly more convincing and likely to pass video verification and other KYC steps. Some existing modules can take a photograph and create a video deepfake to fool less sophisticated liveness checks.||Fraudsters will be able to assume synthetic and stolen identities more easily, thus committing loan fraud, money laundering and dozens of other types of fraud and scams.|
In many cases, solutions are already in place to help mitigate the above. For example, LinkedIn will now remove users who provide AI-generated headshots as legitimate pictures of them, but this only comes after researchers demonstrated that more than 1,000 LinkedIn profile images were created by artificial intelligence.
Fraud Is on the Rise Thanks to Lower Barriers to Entry and the Economic Downturn
Following a trend first seen a few years ago, with Fraud as a Service (FaaS), fraud is about to become more accessible and, thus, even more prevalent – especially fraud that deals with getting money out of the system. The reasons range from economic hardship to the ease of access to sophisticated tools and tutorials for novice fraudsters.
In fact, SEON data shows that the ratio of declined transactions jumped by 10.72% from just October to December of 2022. Once you look at everything that was flagged, inclusive of user actions that were deemed suspicious rather than certainly fraudulent, we are looking at a 12.90% increase in the same metric.
Partner industries that were most affected by this were gambling and casinos (iGaming) as well as neobanks and associated fintech, with the iGaming sector seeing a jump of as much as 25.3% in suspicious transactions and digital banking and neobanks, 8.6%.
The big tech layoffs of 2022 have led to a huge number of tech-savvy individuals becoming unemployed. In the US, more than 58,000 workers are estimated to have lost their jobs – some of whom, statistically speaking, could turn to fraud through need or choice. In anticipation of economic developments in 2023 and beyond, SEON explored the impact of a global recession on fraud with the Global Recession Fraud Report, when we compared historical data by year.
There are several consequences of a recession but more than anything, it makes people more likely to be opportunistic. For example, be they due to first-person fraudsters or organized rings, Frank on Fraud expects this to push up fraud-related defaults “by as much as 50%”. This would involve income fraud, credit washing, employment fraud, loan stacking as well as synthetic identities.
In fact, internal SEON data corroborates this trend, as we have seen a significant uptick in more clever types of fraud indicators. The rules triggered by user actions that are monitored by SEON’s APIs are less obvious and more complex – often related to device data and IP rather than the person’s digital footprint or email address. These include:
- public and web proxies
- privacy extensions
On the contrary, there were much fewer customers signing up with non-registered email address domains, impossible phone numbers, and custom, newly registered domains – what we could label as beginner techniques.
However, we should note that considered in isolation, most of the above rules do not necessarily mean someone is a criminal or attempting fraud. Each of these indicators should be considered in combination with hundreds more data points, which will then create a fraud score that gives a clearer idea of the level of risk involved – to then be actioned by the system or a manual reviewer.
Fraudsters Will Take Advantage of Companies’ Eagerness to Keep Customers Happy
One thing that is clearly evident from SEON data is that the average fraudster is likely to use more than one technologies and tools to conceal their identity and intentions. After all, the better prepared they are, the more likely their schemes will succeed. However, at the same time, businesses are struggling to balance costs and return on investment on their risk software.
It is not just workers who are affected by an economic downturn but companies, too. In addition to having fewer resources to spend – or being less eager to spend them – on prevention due to budget cuts, organizations far and wide are more mindful than ever of churn-inducing customer friction.
Customers have less patience, more options and less disposable income in 2023.
- If a company interrupts the customer experience too many times to double-check login details, verify their identity, and confirm their intentions, it is likely to cause customer churn – and, therefore, lose revenue.
- Similarly, at times of economic hardship, a company can’t risk false positives, which by nature would mean losing the custom of legitimate consumers.
Metrics are important. Organizations are used to calling the shots based on the data, but is it the correct data?
For instance, they’ll think, “the likelihood of any person being a fraudster is very low, so why block them? They’re an opportunity for profit, and we don’t want to miss it.” But in actual reality, one fraudster can cause greater harm than 100 legitimate customers can cause good.
In fact, SEON found that from September to December of 2022, the average amount of money that fraudsters tried to steal per transaction increased by 4x. So, if the average fraudster attempted to steal $100 from you last September, they’re going after $400 on average in December and beyond.
Granted, it is difficult to demonstrate the risk in cold hard numbers because you don’t know the possible income from the customer you’re turning away, or you’re causing to churn. Your boss might say, “You are declining too many people; I want more sales”, so you may start declining fewer people.
Indeed, your aim should be to reduce false positives and customer insult rates without compromising your protection. What you ought to be doing is labeling and studying every action by every fraudster, not just the ones you decline.
What does this mean? When someone is found to be a fraudster later on during their customer journey, make sure you also label their previous actions in the SEON system as fraudulent and share this information with SEON – it will help you to thwart fraud more efficiently.
This will provide you with much more clarity on the harm this person has caused, together with an estimate of money lost or potentially lost. It will also allow you to more easily calculate predicted costs for a fraudster and compare them with estimated gain from a good customer.
By marking everyone and differentiating between fraudulent and non-fraudulent transactions on your system, you will find it much easier to make substantiated predictions and informed decisions rather than going only by instinct or feelings – which may be great in some situations, but not here.
For those deploying fraud detection with machine learning, labeling will help train your ML modules better and faster. On SEON, this will result in improved whitebox ML rule suggestions as well as more accurate blackbox ML scoring.
Fraud Fighters Are Going to Need to Work Together More
This is both a trend and an admonition: This year and in the future, everyone taking a stance against fraud will have to fight closer together than ever to thwart attackers.
By this, I mean both that fraud fighting professionals need now more than ever to work together, and that efficient prevention and mitigation should involve a stack of risk-related solutions and tools instead of one single end-to-end platform – with API functionality enabling better integration.
A good example of this comes from SEON 2022 data that highlighted an increase of 33.6% in the ratio of fraudsters getting caught because their emails were found on blacklists.
What does this mean? Regardless of whether they are in direct contact, those who maintain blacklists are immensely helpful to companies like ours, which are identifying and reducing fraud for customers across different sectors of economic activity. Elsewhere, collaborations take the form of knowledge sharing, partnerships and masterclasses.
The point being, cross-sector collaboration is key. Many in fraud prevention are doing this, but we can pick up the pace.
Anti-fraud solution makers can share some of their knowledge even if they do not share their actual intellectual property or tech stack because they are competing for clients. We can, instead, talk about new trends we are noticing or strategies to stop them. Because fraud is growing in the world.
Fraudsters have their own forums, so why shouldn’t we? Like in the Lord of the Rings, we all need to fight together for a better outcome.
Partner with SEON to reduce fraud rates in your business with real-time data enrichment, whitebox machine learning, and advanced APIs.
Book a Demo
The Adoption of Anti-Fraud ML Is about to Skyrocket
The use of machine learning for fraud prevention will continue to increase in 2023, as more companies enlist its help to reduce risk.
What does this mean? That decision-makers will become more likely to invest in fraud prevention solutions that integrate machine learning modules, as well as pay more attention to any ML output from the solutions they already have in their stack.
At SEON, since September 2022, the use of whitebox machine learning rule suggestions has increased by over 30%. The whitebox ML module generates new rule suggestions based on the organization’s historic fraud data, and kicks in after at least 1000 transactions have been logged.
In that same time period, SEON’s blackbox machine learning module – which provides a separate fraud score to inform decision-making – is now being used 46% more.
Figures across the industry confirm this trend both for fraud and adjacent concerns, such as identity verification and AML. For example, SAS found that 33% of banking institutions accelerated their adoption of machine learning in anti money laundering software over the pandemic, with the primary driver being to “improve the quality of investigations and regulatory filings”.
To boost your protection while using machine learning, consider these pieces of advice:
- Ensure you always label fraudulent user actions properly, to boost accuracy.
- Blackbox fraud scores will be able to catch suspicious patterns never seen before, making it a great line of defense against new trends in fraud – but are not easily explainable.
- Where possible, use both whitebox and blackbox ML to get the best of both worlds: Human intelligence and ML-human collaboration with SEON’s whitebox module and emerging new trends and attack vectors with SEON’s blackbox fraud scoring.
For the above figures, we have used internal SEON data aggregated from our 99 most relevant and biggest in terms of data customer accounts, collected from December 2021 to January 2023.
- About Fraud: Dawn Of The ShapeShifter – 10 Fraud Predictions for 2023
- CSO: Study shows attackers can use ChatGPT to significantly enhance phishing and BEC scams
- Crunchbase: Tech Layoffs: U.S. Companies That Have Cut Jobs In 2022 and 2023
- Silicon Republic: Researchers find 1,000 LinkedIn profiles using computer-generated faces
- SAS: Pandemic spurs banks’ AI adoption for AML, SAS study shows
Showing all with `` tag
How to Detect Payment Fraud in Buy Now, Pay Later
How to Prevent Cryptocurrency Account Takeover
AML & CFT: Combating Money Laundering & Financial Terrorism Financing
What You Need to Know About KYC for Online Lending
See a live demo of our product
Tamás Kádár is the Chief Executive Officer and co-founder of SEON. His mission to create a fraud-free world began after he founded the CEE’s first crypto exchange in 2017 and found it under constant attack. The solution he built now reduces fraud for 5,000+ companies worldwide, including global leaders such as KLM, Avis, and Patreon. In his spare time, he’s devouring data visualizations and injuring himself while doing basic DIY around his London pad.
Sign up for our newsletter
The top stories of the month delivered straight to your inbox