What is missing in the fight against AI-enabled fraud

Gal Bello at Neo4j argues that connected-data approaches are becoming essential because AI-enabled fraud often leaves patterns across networks, even when individual events appear legitimate

The nature of modern fraud is almost unrecognisable to the prior tactics, patterns, and behaviours that surfaced during the online boom. Bad actors are no longer ‘trying their luck’ with unsophisticated attempts; instead, fraud is driven by mass organised crime networks that are highly sophisticated, coordinated, and operate at scale.

At the same time, the rapid development of AI has increased the volume of fraud. Organisations now face around-the-clock threats across every channel. This comes at a point when fraud education and awareness are at their highest, driven by public interest in AI-created threats, like deep fakes. Attacks are now likely to be highly personalised, catching unwitting employees off guard, and causing confusion and business vulnerability.

In this environment, everyone is at risk. Legacy protection tools are not enough. Businesses need to be proactive in the fight against fraud and have advanced, robust support that can identify illicit behaviour and spot fraud patterns before the ripple effect is felt across entire operations.

The fraud networks hiding in plain sight

In the UK, fraud now accounts for more than 44% of all crime, making it the single largest category of criminal activity. In the first half of 2025 alone, £629.3 million was lost to payment fraud and scams, with businesses and individuals both falling victim.

It’s the complex web of modern fraud that makes it difficult to detect. Businesses are finding that detection and prevention are not linear; mule accounts are linked to synthetic identities, compromised credentials are reused across platforms, and shared devices connect profiles that appear unrelated. What’s more, activity is deliberately fragmented across multiple accounts, identities, and systems so that no single step appears unusual, exploiting the gaps between disconnected systems.

The challenge for organisations is no longer simply identifying suspicious activity but understanding how it all fits together in order to keep it out.

In order to achieve such a wider perspective on all threats, accounts, identities, activities, transactions, we need to gather all data entities into a centralised data platform that can assist us identify the threats and act accordingly. This is where Graph Database Platforms & GraphRAG enters.

Why fraud detection needs a new approach

More than half of fraud cases now involve AI-enabled tactics, such as synthetic identity generation and phishing scams, making fraudulent activity looks increasingly legitimate and threats more difficult to trace. Legacy fraud detection relies on traditional relational databases that analyse data in isolation. While these systems were designed with rule-based controls to detect individual anomalies, they are unable to interpret fluid, AI-fraud networks that are constantly advancing and adapting.

Graph databases have become an imperative for businesses wanting to protect themselves against fraud. By mapping data relationships across people, devices, identities, and events, graph intelligence technology gives organisations the context needed to uncover hidden connections and coordinated fraudulent activity, by providing a Data Model that maps and connects all entities which will serve to understand the entire technical stack.

iFood, the largest digital delivery platform in Latin America, Graph technology helped fraud teams uncover organised promotion abuse by revealing hidden relationships between Users, Restaurants, Devices, and Payment credentials. Within a week of using Graph Data Platform combined with GraphRAG processes, the iFood team had built six fully functional fraud detection use cases.

Connected data, better decisions

With many organisations now holding vast amounts of customer, operational, and regulatory data to operate effectively and detect suspicious activity, businesses are discovering that this information is often fragmented, living across siloed systems and business functions.

This means that when AI is introduced for fraud detection or other processes, it may lack the necessary context to produce meaningful, accurate outputs. Instead, it can hallucinate – providing biased, false, or misleading insights – and cause unnecessary delays. In fraud, it could mean false alerts, wasted resources on manual case resolution, and even bad actors passing through unchecked.

By connecting data and providing richer context within a knowledge layer, graph technology improves the quality, transparency, and reliability of AI outputs, creating a stronger foundation for confident decision-making.

Addressing the AI – and data – problem

Criminals will continue to embrace new technologies to accelerate fraud attacks and make them harder to detect. But organisations need to look beyond simply keeping pace with the latest tactics. Instead, they need to address the underlying issue that will allow them to catch fraud activity: disconnected data.

AI is only going to become more powerful and sophisticated. Therefore, the organisations best placed will be those that integrate their data to enable faster, more informed and data driven decisions. Connected data within a knowledge layer should be viewed as a strategic asset, helping organisations move from reacting to individual incidents to identifying coordinated fraud before it causes significant damage.