FB pixel

AI may be creating new fraudsters, not just enabling existing ones: Socure

Research suggests AI is enabling new fraudsters, exposing the limits of traditional identity verification
AI may be creating new fraudsters, not just enabling existing ones: Socure
 

Generative AI may be expanding the population of people willing to commit fraud, rather than simply giving established criminals more powerful tools, according to new research from Socure.

The identity verification and fraud prevention company found that 13.4 percent of respondents in a global consumer survey reported using AI to commit at least one type of fraud. Those classified as AI-fluent fraudsters attempted twice as many types of fraud and targeted 40 percent more businesses than respondents who reported committing fraud without AI.

Socure estimates that the combined frequency and breadth of their activity produces roughly six times the fraud volume of traditional offenders. More than half of AI-fluent fraudsters reported committing fraud at least 10 times, compared with 9.2 percent of traditional fraudsters.

The findings raise the possibility that AI is lowering barriers to fraud among consumers who might not otherwise engage in it. However, the self-reported survey cannot establish that AI caused respondents to begin committing fraud.

The research appears in Fraud’s Napster Moment, the inaugural edition of Socure’s From the Frontline Fraud Insights series, led by Mike Cook, the company’s head of fraud insights and co-founder of ID Analytics.

Cook argues that generative AI is doing for fraud what file-sharing technology did for music piracy: making capabilities once largely associated with organized criminal operations accessible to ordinary consumers.

AI changes the fraud landscape

Cook has spent nearly 40 years tracking fraud. He frames the report as a call to arms. In his foreword, he writes that fraud has accelerated sharply in just the past six months, and that the way criminals build identities has “changed completely.” The gibberish email addresses and obvious red flags fraud teams once relied on are, in his words, long gone.

What has replaced them is a criminal economy that learns and collaborates. Bad actors buy services from one another, share techniques, and trade stolen identities, deepfakes and ready-made synthetic identity packages. AI is used to build fake identities with realistic digital footprints, automate attacks at scale, and produce increasingly convincing deepfakes and counterfeit documents. The offender can be a sophisticated criminal, a nation-state operative, an ordinary consumer or a rogue AI agent. Socure has labeled the period “World War Fraud,” a conflict in which the battlefield shifts quickly and once-exclusive tools are open to almost anyone.

Research draws on survey and fraud network data

The first issue of the report draws on two sources.

The first is Socure’s identity and fraud consortium. The company says it holds more than 20 billion transactions, 620 million accounts and 178 million unique identities, which it puts at about 72 percent of the adult banking population in the US. The data comes from more than 3,000 customers across 30 industries and more than 190 countries. Socure flags fraud using high score thresholds of 0.975 to 0.99 in its Sigma models, so it describes its growth and volume figures as conservative.

The second is a global consumer survey run through the Prolific platform, with 1,253 confirmed human respondents after quality checks. Respondents counted as AI-Fluent Fraudsters if they reported at least one AI-assisted fraud behavior occasionally or more often, across first-party fraud, synthetic identity fraud and identity theft. Traditional Fraudsters reported manual fraud only, and the rest of the sample met neither definition.

The report also sets the wider scene. It cites an INTERPOL estimate of $442 billion in global financial fraud losses in 2025. On Socure’s own network, deepfake attacks rose 8,000 percent year over year in 2025, identity fraud risk climbed about 134 percent between June 2024 and June 2026, and first-party fraud rates rose 20.7 percent over the last six months.

The new face of fraud

Socure’s profile of this group is a Gen Z male. AI-Fluent Fraudsters averaged under 30 years old, compared with 33.8 for traditional fraudsters and 35.5 for non-fraudsters. Nearly 63 percent were male, against 55 percent of traditional fraudsters.

The AI-fluent group is also twice as likely to go ahead with a 50/50 chance of being caught. Socure suggests one reason for their greater frequency of fraud is reinforcement: when a fraudulent claim succeeds, the person learns what the system will accept, gains confidence and repeats the tactic elsewhere, a cycle AI makes easier to start and to scale.

Financial hardship is not the leading motivation among AI-Fluent Fraudsters surveyed. Curiosity was the most cited motivation at 41.8 percent, ahead of economic hardship at 37.7 percent, and 22.3 percent said knowing someone who had committed fraud played a part. AI also appears to soften the sense of guilt. About 16.5 percent of AI-Fluent Fraudsters said using AI did not really feel like committing fraud, compared with 6.1 percent of traditional fraudsters. The report argues that just as Napster made stealing music feel invisible, AI makes fabricating documents and explanations feel outsourced.

Socure’s sharpest claim is that AI is creating fraudsters, not just equipping existing ones. Strip out AI-related behaviors and the two groups look statistically the same, at 4.1 percent versus 5.0 percent. More than half of AI-Fluent Fraudsters showed no evidence of fraud without AI, and many said they would not know how to commit the behavior without it.

The distinction matters. If AI merely made existing fraudsters more productive, organizations would face a problem of increasing attack volumes. If it is also enabling people who would not otherwise commit fraud, the industry faces a potentially expanding population of offenders. The survey raises that possibility, although it cannot establish causation.

First-party fraud as the gateway

The report identifies first-party fraud as the easiest way in, because it needs no new or stolen identity. A person can use their own name and account while submitting fabricated evidence. Among AI-Fluent Fraudsters, 81 percent engaged in first-party fraud, 44.6 percent in synthetic identity fraud and 13.6 percent in identity theft. Some 42.3 percent were active in two or more fraud types at once, and respondents aged 25 to 34 committed 2.83 times more first-party fraud than everyone else.

The gap with traditional fraudsters is widest where convincing paperwork is needed. Some 61.3 percent of AI-Fluent Fraudsters had used fake hardship letters, compared with 16.6 percent of traditional fraudsters. For fake income or bank statements, the figures were 32.1 percent and 8.6 percent, and for skipping a buy now, pay later (BNPL) payment, they were 51.2 percent and 24.5 percent. The report also cites a Liminal survey in which 80 percent of card-not-present fraud leaders said they had received dispute evidence they believed was AI-generated or fabricated.

Part of the appeal is that first-party fraud is easy to justify as recovering money one feels owed, which lowers the psychological barrier to trying it in the first place.

Following the incentives, and the rings

AI-Fluent Fraudsters over-index where sign-up bonuses drive customer acquisition. They target gaming and gambling platforms at nearly three times the rate of traditional fraudsters (31.7 percent versus 10.9 percent), BNPL providers at more than double (26.7 percent versus 12.3 percent), and telecom carriers at about 2.5 times (21.7 percent versus 8.7 percent). Gambling is especially attractive because sign-up bonuses and free bets can pay out again with every new fake identity, turning AI’s ability to mass-produce identities into repeatable profit.

Organized rings amplify the problem. Survey respondents who coordinated with others were far more likely than lone actors to use AI (67.3 percent versus 42.9 percent), to commit fraud frequently (54.5 percent versus 23.6 percent) and to work across two or more fraud types (50.9 percent versus 23.2 percent). Lone actors mostly cited hardship, while ring members pointed to curiosity and social influence.

The report describes a “confidence without fear” mindset among both groups. Many believe companies arm fraud teams with the latest AI detection tools. They simply assume the consequences will not reach them personally.

The World Cup as a live test

To show the behavior in practice, Socure examined online sports betting and prediction-market activity during the 2026 World Cup and compared it with a 30-day pre-tournament baseline. It tracked 451,884 fraud attempts tied to the tournament. Daily attempts rose from a baseline of 3,083 to 11,587, and high-risk activity grew 303.9 percent from the first day to the final while transaction volume grew 154.1 percent.

The attacks followed the money. They clustered around sign-up bonuses, referral rewards, promotional credits, deposit offers and fast onboarding flows. Fraudsters combined real and synthetic identities with burner phone numbers, disposable domains, proxy infrastructure and coordinated account creation. About 90 percent of attacks used real consumer identities.

The finding is significant for identity verification providers. Even when an identity is genuine, the associated activity may be fraudulent. It highlights the distinction between verifying identity authenticity and determining whether a transaction or account activity is legitimate.

Socure argues that point-in-time checks alone would struggle to detect these patterns, because the attacks moved across identities, devices, domains and accounts as the opportunity moved.

What it means for buyers and vendors

The report argues that identity verification is the starting point, not the finish line. Its closing question captures the shift. Fraud teams should stop asking only whether an identity is real, and start asking whether the person, document, claim, device and behavior can be trusted together.

For organizations buying fraud and identity tools, the first lesson is that a static, one-time check at onboarding no longer holds up against an adversary that keeps adapting. Document validation alone is not enough either. The stronger approach builds a 360-degree risk profile of each person from many signals, including identity, document, behavioral, device and velocity data, and checks whether those signals connect across institutions and industries. Defenses also need to cover the full spectrum, from first-party fraud through synthetic identities to identity theft. These tools cannot be deployed and forgotten. Buyers need to keep updating them as attacks evolve, and measure their performance continuously rather than at procurement time alone.

The findings also highlight the boundaries of biometric identity verification and authentication. Facial matching, liveness detection and document verification can help organizations establish identity authenticity and detect impersonation attempts. But they cannot, on their own, determine whether a legitimate customer is submitting a fabricated hardship claim, falsely disputing a transaction or exploiting a promotional incentive. These scenarios require additional fraud and behavioral intelligence.

Reading the numbers

The survey findings come with limits that Socure itself acknowledges. Responses are self-reported. The methodology also states that the survey shows relationships within the sample but cannot establish the order of behaviors or prove on its own that AI caused additional fraud. The AI-fluent group is about 13 percent of 1,253 respondents, or roughly 170 people, so the subgroup percentages rest on a modest base.

The consortium data provides additional evidence of rapidly evolving fraud patterns, although it measures different activity from the consumer survey. The report’s broader warning is that this is not only a technology problem but potentially a behavioral and social one. AI is lowering the effort required to commit fraud, while some users appear less likely to recognize AI-assisted deception as wrongdoing. Whether this translates into a lasting shift in consumer behavior remains uncertain.

For fraud prevention organizations, the immediate challenge is clear: verifying that someone is real is no longer enough to establish that their actions are legitimate.

Related Posts

Article Topics

 |   |   |   |   | 

Latest Biometrics News

 

Making digital identity work takes more than technology

European analyses suggest biometric technologies can be used effectively without compromising privacy, while one of the EU’s major projects takes…

 

North Korean hackers steal faces for crypto fraud and cyberattacks

North Korean hackers have turned hundreds of real people’s faces into reusable tools for cyberattacks, secretly recording victims during counterfeit…

 

Privacy-enhancing technologies boom, but adoption lags: OECD

The Organisation for Economic Co-operation and Development (OECD), the Paris-based policy body of 38 advanced economies, has published a new…

 

Google’s AI glasses raise questions about enforcing biometric privacy protections

Google is entering the AI glasses market with privacy protections designed to address controversies surrounding Meta’s smart eyewear. But its…

 

Biometric defenses put to the test with faster AI face-swap model

The best way to stop a face-swap attack may be to build a better one first. Sumsub and Constructor University‘s…

 

Podcast: EU and UK online safety laws keep coming. Are they making the internet safer?

Online safety regulations are reshaping the internet as governments introduce new measures to reduce online harms. Two of the biggest…

Comments

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.

Market Intelligence

Biometric Update Podcast

Featured Company

Top Stories This Week

White Papers

Latest Webinars

Biometrics Industry Events