AI is industrialising E-Commerce Fraud

Artificial intelligence is transforming online fraud from a collection of individual scams into something resembling an industrialised criminal supply chain, allowing attackers to operate at unprecedented speed and scale.

AI is industrialising E-Commerce Fraud

New research from Signifyd finds overall E-Commerce fraud pressure increased 33% year on year during the first four months of 2026, while first-party fraud and abuse rose 9%. The company defines fraud pressure as the proportion of orders on its network considered highly risky.

More striking is what is happening beneath the headline figure. Account takeover attempts increased 78% year on year between January and April, while card-testing attacks — used to establish whether stolen payment credentials remain active — soared 175% during the first four months of the year.

AI Changes the Economics of Payment Fraud

Card testing illustrates the shift particularly clearly.

Rather than manually attempting stolen credentials, automated bots can bombard ecommerce sites with thousands of small transactions an hour. Successful cards can then be classified and prioritised before criminals use them for higher-value purchases. The process effectively turns stolen payment data into a rapidly tested and ranked inventory.

AI is similarly changing account takeover. Fraudsters can automate credential attacks, create convincing imitation retail sites and generate personalised phishing campaigns at a scale that would previously have required substantial criminal organisations.

Signifyd argues that online fraud should consequently be viewed as an interconnected ecosystem spanning identity theft, account takeover, stolen cards, synthetic identities, fraudulent purchases and refund abuse.

When Photographic Evidence Can No Longer Be Trusted

Generative AI is also creating a different problem after the payment has been completed.

Fraudsters can manufacture realistic receipts carrying genuine retailer branding or generate images apparently showing products arriving damaged. Signifyd cites examples ranging from supposedly shattered glassware to damaged bedding and household products.

The economics are significant. US retail returns reached $850bn in 2025, of which $76.5bn were fraudulent. Across Signifyd’s wider dataset, claims that online products arrived “not as described” increased 49%.

European Merchants Face a Similar Escalation

The problem is equally visible across Europe and the UK.

Signifyd recorded a 67% increase in account takeover attempts during the first third of 2026. Claims that packages had not arrived increased 49%, while “not as described” claims rose 35%.

Risk is also shifting between retail sectors. Fraud pressure increased 161% for grocery and household goods, 97% for consumer medical supplies and supplements, 86% for leisure and outdoor retailers and 55% for electronics.

Fraud Prevention Must Follow the Entire Payment Journey

The implications extend beyond deploying better fraud models at checkout.

AI is reducing the cost of discovering vulnerabilities, validating stolen credentials, impersonating customers and manufacturing convincing evidence. Fraud prevention therefore increasingly needs to connect signals across the entire transaction lifecycle — identity, account behaviour, payment, fulfilment, refunds and returns.

The contest is no longer simply about whether AI can detect fraud.

It is whether defensive AI can identify it faster than offensive AI can manufacture it.