AI did not invent identity fraud. It changed the economics of it. For most of the last decade, running a convincing fraud operation took time, skill, and money, which kept the volume of serious attacks in check. AI fraud removes those constraints. A deepfake that once required a specialist now takes minutes and a cheap model. Synthetic identities that took weeks to build can be produced at scale. The result is not simply more fraud, but a different kind: faster to launch, cheaper to repeat, and convincing enough to clear checks that used to stop it. In the report Liminal recently co-authored with Unico, the cost of running a sophisticated attack has fallen more than 100x. When the price of an attack drops that far, everything downstream of it changes.
What makes AI fraud different
AI fraud is any fraud scheme that uses artificial intelligence to create, scale, or disguise an attack. In identity fraud specifically, it shows up in three ways.
The first is fabrication. Generative models produce synthetic identities, forged documents, and deepfake audio and video that pass as real. Group-IB tracked $347M in losses across 8,065 deepfake-enabled attempts, a category that barely existed a few years ago. The UK government projects the number of deepfakes in circulation will reach 8 million in 2025, up from 500,000 in 2023, a roughly 16x rise in two years.
The second is scale. What used to be a manual, one-at-a-time craft is now automated. An operator can generate thousands of variations of an identity and test them against onboarding systems until some get through.
The third is evasion. AI-generated attacks adapt to the defenses in front of them, which is why 23.3% of fraud now sits in what the report classifies as a sophisticated tier, built specifically to defeat detection.
The economics flipped
The headline shift is economic. When an attack costs 100x less to run, fraud stops being a series of discrete incidents and becomes a continuous, industrial activity. The returns justify constant attempts, and the low cost means a failed attack barely matters, because the next one is nearly free.
The loss figures follow. Industry estimates put global fraud losses past $400B a year, and Liminal’s research projects financial-institution losses growing 121% by 2030, reaching $55.3B. Independent forecasts point the same way, with Juniper Research projecting fraud will cost financial institutions $58.3B by 2030. This is not a linear increase on top of existing fraud. It is the compounding effect of attacks that are cheaper to run and harder to catch at the same time.
The confidence gap
Here is the uncomfortable part. Most teams do not feel exposed. In Liminal’s research, 93% of practitioners say they are confident in their fraud models. At the same time, 92% acknowledge that legacy infrastructure still lets fraud signals slip through. Both numbers are true at once, and together they describe the central risk of this moment: high confidence resting on aging plumbing. The wider industry sees the same tension, with the World Economic Forum noting that overall fraud rates edged down from 2.6% to 2.2% even as attack complexity climbed.
The models are not the problem. Detection has improved. The gap is in the infrastructure underneath, which was built for a slower, more expensive kind of fraud and now has to contend with attacks that arrive faster and look more legitimate than those systems were designed to expect.
Latin America is the preview
To see where this is heading, look at Latin America. In the markets the report examines, synthetic identity now accounts for 48.3% of fraud, a level most other regions have not reached yet. That is not because the region is uniquely vulnerable. It is because adoption of digital identity moved fast, and fraud moved with it.
Latin America is a preview, not an exception. The same forces, cheaper attacks, better fakes, and faster digital onboarding, are present everywhere. Other markets are simply at an earlier point on the same curve. Treating the region as a leading indicator rather than an outlier is the more useful posture.
What connected defense requires
If AI made attacks cheaper and more coordinated, the response has to be more coordinated too. The pattern in the data is consistent: defenses that share context outperform defenses that operate alone.
Buyers already sense this. 76% of practitioners now rank device fingerprinting as their most effective defense against AI-enabled fraud, a move toward signals that persist and adapt rather than one-time checks. Datos Insights finds that firms adopting behavioral biometrics and device fingerprinting report fewer false positives and stronger detection of account takeover. Strengthening partnerships is the fastest-rising investment priority for 2026, named by 47% of buyers, because no single institution sees enough on its own.
The reason isolation fails is visible in the network data. In the report, a single fraud entity was tied to 949 distinct identity documents, and one operator was seen targeting as many as 30 different businesses. Each of those institutions met that actor for the first time, because none of them could see what the others already had. Connected defense closes that gap by treating identity, authentication, and fraud as one system with a shared memory, so a signal caught in one place is available everywhere else.
Frequently asked questions
What is AI fraud?
AI fraud is any fraud that uses artificial intelligence to create, scale, or disguise an attack. In identity fraud, that includes deepfake audio and video, AI-generated synthetic identities, and automated attacks that adapt to the defenses they encounter.
How do deepfakes enable fraud?
Deepfakes let attackers impersonate real people convincingly in video, voice, and documents, which defeats verification steps that assume the person on the other end is genuine. Group-IB tracked $347M in losses across 8,065 deepfake-enabled attempts.
What is synthetic identity fraud?
Synthetic identity fraud combines real and fabricated information to create an identity that does not belong to any single real person. AI has made these identities faster to produce and harder to tell apart from genuine ones. In the markets Liminal’s report examines, synthetic identity accounts for 48.3% of fraud in Latin America.
How do you defend against AI fraud?
The most effective approaches connect identity, authentication, and fraud signals so they share context in real time, rather than relying on isolated point tools. 76% of practitioners rank device fingerprinting as their top defense, and 47% name strengthening partnerships their leading 2026 priority.
Key takeaways
- AI did not create more of the same fraud. It changed the economics, making attacks cheaper to run, faster to repeat, and convincing enough to clear existing checks.
- The cost of a sophisticated attack has fallen more than 100x, and financial-institution losses are projected to grow 121% by 2030 to $55.3B.
- There is a confidence gap. 93% of teams trust their fraud models while 92% admit legacy infrastructure still lets signals slip through.
- Latin America, where synthetic identity is 48.3% of fraud, is a preview of where other markets are heading, not an exception.
- Connected defense, where identity, authentication, and fraud share context in real time, is where buyers are moving: 76% rank device fingerprinting their top defense and 47% name partnerships their top 2026 priority.
AI has reset the economics of identity fraud, and the markets feeling it first are showing everyone else what comes next. To see the full picture, including how attack costs collapsed and where connected defense is working, read Identity Fraud Intelligence: Trends and Insights, the report Liminal co-authored with Unico. Download the report
