AI is making identity fraud cheaper, faster, and easier to scale, putting pressure on e-commerce companies to rethink identity checks performed at a single point in time.
Fraud prevention and identity verification firm Microblink found that attackers are increasingly using AI to alter legitimate identity documents rather than creating convincing fakes from scratch. Its analysis also found that portrait forgery, in which an attacker replaces or manipulates the photograph on an otherwise legitimate identity document, accounted for roughly 60% of document fraud failures observed in the U.S. and nearly 65% in Canada.
Microblink’s approach combines document authentication, biometric facial matching, liveness detection, and other risk signals to assess identity throughout a customer’s interaction with a business rather than relying solely on an initial verification.
“Fraudsters no longer need to build convincing fake identities from scratch. They can modify legitimate credentials quickly, making traditional verification approaches increasingly difficult to rely on,” Microblink CEO Hartley Thompson III told the E-Commerce Times.
Moving Beyond One-Time Verification
Microblink argues that identity verification needs to draw on multiple signals rather than operate as a standalone check. Its research found that fraud techniques vary significantly by geography, suggesting that defenses designed around a single attack pattern may miss threats that are more prevalent in particular markets.
The company’s technology combines document authentication, payment card fraud signals, biometrics, and liveness detection. Its Fraud Lab also develops and tests defenses against synthetic identities, manipulated documents, and deepfakes.
Microblink’s verification process involves four steps:
- Users scan government-issued IDs via mobile or web cameras, guided by real-time SDK feedback for optimal image quality
- AI models check visual security features, inspect barcodes and MRZ lines, and spot tampering, forgery, or generative AI manipulation
- Biometric facial recognition and liveness detection confirm that a live user matches the portrait on the scanned ID and is not using a screen replay or deepfake
- Automated risk scoring generates instant decisions to block bad actors while letting real customers pass
AI Changes the Forgery Equation
Thompson described what Microblink calls the “sophistication paradox,” in which stronger physical security features on government IDs drive fraudsters toward precise AI-based editing of legitimate credentials rather than total fabrication.
“The paradox is that the more secure legitimate documents become, the more incentive there is for fraudsters to manipulate a real document rather than create a fake one from scratch,” he said.
Microblink’s analysis of millions of identity interactions during the first half of 2026 found substantial regional differences in attack methods. Portrait forgery was more prevalent in North America. Screen presentation attacks were more common in Europe, whereas physical replicas were more frequent in Latin America, the Middle East, and Africa.
The pattern suggests fraud operators are adapting their methods to local identity documents, infrastructure, and verification controls rather than relying on a single technique worldwide.
“AI has made that manipulation much easier. An attacker can take a legitimate credential, change the photograph or other fields, and preserve enough of the document’s original structure, layout, barcode, and security features to make it look legitimate. That’s a very different problem from detecting a crude counterfeit,” Thompson said.
Seeing Is No Longer Believing
Visual inspection alone is becoming less reliable, Thompson said. Verification systems increasingly need to determine whether the information embedded in a document is consistent with what appears on its face.
“One example we see is a barcode anomaly where the encoded information doesn’t match what is visibly displayed. The document can look essentially perfect to a person, but the underlying data tells you something is wrong,” he explained.
Thompson urged organizations to watch not only for increases in fraud volume but also for changes in the types of attacks they encounter.
“You may suddenly see a concentration of a particular document attack, a new manipulation technique, or activity that looks very different from what you’ve historically seen in that market. That can be a sign that attackers have found something that works and are beginning to scale it,” he said.
The Limits of One-Time Verification
Many identity systems concentrate verification at onboarding or another discrete point in the customer journey. The weakness of that model is that a legitimate verification does not guarantee that an account or transaction will remain trustworthy later.
Accounts can be compromised after onboarding, while synthetic identities can accumulate enough history to appear increasingly legitimate. Changes in devices, behavior, transaction patterns, or other signals can also alter the risk associated with an identity after the initial check.
Other fraud researchers are seeing a similar need to evaluate risk across the customer lifecycle. Sift‘s Q2 2026 Digital Trust Index found that fraud rings can be difficult to detect when accounts, devices, payment instruments, and transactions are examined separately. The company said stronger fraud programs increasingly connect identity, payment, behavioral, device, and outcome data to identify coordinated attacks.
Sift also found that the average cost of a false positive in digital commerce reached $496 in the first quarter of 2026, underscoring the potential cost of fraud controls that mistakenly block legitimate transactions.
Thompson said continuous assessment does not necessarily mean repeatedly asking customers to verify their identities. Instead, systems can reassess risk using signals such as device intelligence, behavioral patterns, biometrics, transaction context, and document authenticity as interactions continue.
GenAI Lowers the Cost of Fraud
The financial impact of AI-assisted identity fraud is difficult to quantify, but its ability to automate attacks changes the economics for both fraudsters and the businesses trying to stop them.
Thompson said generative AI has made some fraud techniques faster, cheaper, and easier to scale by reducing the expertise and manual effort needed to manipulate documents or create synthetic media.
“Once you can automate the process, you’re no longer talking about a handful of attempts. You’re talking about systems that can generate identities, test them, learn from what works and keep going around the clock,” he warned.
