Deepfake Fraud Surges Over 2,000% in 2 Years, Reshaping Fintech Security

Previously, the biggest threat in the online banking world was relatively predictable.

Fraudsters would typically steal a password, a credit card number, or other sensitive credentials before they could access someone’s account.

In other words, they needed a real identity to exploit. But this is no longer the case.

Alexey Astakhov is the Vice President of Engineering at Instinctools, a leading AI-powered software engineering company.

He says that as generative AI tools become easier to access, criminals are finding ways to bypass identity verification systems altogether.

“Instead of stealing credentials, they can now generate convincing synthetic faces, manipulated ID documents, and biometric deepfakes that are designed to pass digital onboarding checks,” Astakhov says.

“For financial institutions that rely on remote customer verification, the implications are becoming harder to ignore, forcing them to rethink how digital identity verification is conducted in a secure environment.”

The video below illustrates why a major Australian bank recently sounded the alarm over scammers using celebrity deepfakes to target customers.

Editor's Note: This is a sponsored article created in partnership with Instinctools.

Deepfake and AI Fraud Are Surging Across Digital Banking

The problem is growing quickly as generative AI tools become easier to use.

Research from Signicat shows overall fraud attempts rose 80% between 2023 and 2025, while deepfake fraud surged 2,137% over the same period.

These attacks now account for 6.5% of identity fraud cases globally, and highlight how quickly synthetic identity manipulation is entering the financial system.

What makes deepfake fraud particularly dangerous is how convincing it can be.

Manipulated ID documents or synthetic biometric images can appear legitimate enough to pass both automated verification systems and human reviews.

“For digital banks designed to onboard customers quickly, this vulnerability creates a new layer of risk,” Astakhov says.

Take a look at how easily fraudsters can use publicly available information and AI tools to create deepfake video calls that are difficult to distinguish from reality:

Deepfake Fraud by the Numbers

Industry data suggests that the threat extends far beyond isolated incidents, and the pace of change has accelerated sharply in just a few years.

Researchers estimate the number of deepfake videos and images online expanded from roughly 500,000 in 2023 to more than 8 million last year.

These numbers show how rapidly the tools used to create synthetic identities are spreading.

Organizations are already encountering the consequences.

According to Regula’s 2025 Identity Fraud by Numbers report, nearly 50% of companies have experienced fraud involving deepfake audio or video. As a result, 66% of business leaders now consider deepfakes a serious threat.

And consumers are also feeling the impact.

The same report shows that 85% of US consumers believe AI makes scam detection more difficult. And 62% say they’ve either experienced an AI-driven scam or know someone who has.

Security is also a deciding factor for financial institutions competing for customer trust.

The findings indicate that 97% of consumers say fraud prevention is a key factor when choosing a bank. This increases pressure on fintech companies to strengthen identity verification as AI-driven fraud rises.

“These numbers show that attackers are no longer just stealing identities, but manufacturing them for their illicit benefit,” Astakhov said.

Why Fintech Companies Are Turning to Custom AI Fraud Detection

Many financial institutions now recognize that traditional verification tools were not built for the era of generative AI.

These legacy systems still rely on rule-based checks or simple biometric comparisons.

“Deepfake technology can bypass these safeguards by generating synthetic images that appear legitimate to standard detection models,” Astakhov said.

Custom software development offers a more adaptive approach.

Machine learning models can detect the subtle signals that manipulated images often leave behind, such as pixel inconsistencies, irregular lighting patterns, and facial structures that simply don’t align.

“While these details are almost invisible to the human eye, algorithms trained on fraud patterns can identify them quickly,” Astakhov said.

“And because these models continuously learn from new fraud attempts, they can evolve alongside the tactics used by attackers.”

For fintech companies operating in fast-moving digital markets, that flexibility is becoming essential.

Detecting Deepfake IDs in Digital Banking

Instinctools worked with a digital bank in the Netherlands, handling close to five million applications a year.

As volumes increased, so did the chances of fraudulent documents slipping through the customer onboarding process.

And organizations have already felt the consequences.

According to Regula’s 2025 Identity Fraud by Numbers report, nearly 50% of companies have experienced fraud involving deepfake audio or video.

For this reason, the bank needed a way to catch fraudulent documents without adding friction for legitimate users.

Instinctools developed a solution.

Its team built a computer vision system that looks for subtle signs of manipulation in ID images. It doesn’t rely only on standard checks that assess whether a document appears valid.

The system wasn’t designed to replace the bank’s existing verification platform. Instead, it runs alongside it, with both systems reviewing documents simultaneously.

Using a YOLO-based computer vision pipeline, the system identifies irregularities across key parts of a document and assigns a fraud risk score.

Depending on that score, applications can be approved automatically, flagged for extra checks, such as selfie or NFC verification, or sent for manual review.

“The system has been highly successful, detecting 30% more AI-generated fraud attempts, while reducing false positives by 60% and lowering the need for manual verification reviews by 40%,” Astakhov said.

“The result was a stronger onboarding process that improved fraud detection while maintaining a smooth experience for legitimate customers.”

Instinctools has had great success in not only the fintech industry but in other industries as well, as evident in the case study below:

What Fintech Leaders Must Do to Stop Deepfake Fraud

As experts in the field, Astakhov notes that Instinctools approaches deepfake fraud in three ways.

Firstly, fraud techniques are evolving too quickly for rigid systems to keep pace. This means that identity security can no longer rely on static verification methods.

Secondly, tailored fraud detection tools are becoming increasingly important. Systems designed specifically for a bank’s onboarding process can identify patterns that generic software may overlook.

Lastly, security solutions must still protect the customer experience. Overly aggressive verification systems can frustrate legitimate users and slow growth.

“For fintech leaders, the challenge is balancing speed, security, and trust,” Astakhov said.

The United States Senate Federal Credit Union outlines how AI scams work and how to discern deepfakes from the real thing:

The Future of Identity Verification

For financial institutions, the challenge is no longer simply detecting stolen credentials, but having the means to verify if a digital identity is real in the first place.

“In practice, this means investing in stronger identity verification frameworks, integrating machine learning-based fraud detection, and working with specialists who understand how synthetic media and AI-generated identities behave,” Astakhov said.

For fintech leaders, the goal is not only to stop fraud but to preserve customer trust in a digital ecosystem where identity itself is becoming increasingly fluid.

Because in the age of generative AI, the greatest risk to digital banking may not be stolen credentials at all, but the moment when a fake identity becomes indistinguishable from a real one.