Beat the growing fraud challenges of generative AI

Until recently, creating new identities and refining methods to commit fraud was a time-consuming process, requiring extensive research to piece together new synthetic identities, or to groom and socially engineer consumers. Now, though, AI allows fraudsters to fabricate thousands of new synthetic IDs for fraud attacks extremely rapidly using a combination of real and fake consumer information.

As an additional challenge, fraudsters can use generative AI to create fake identity documents that are virtually identical to official IDs, right down to the position of folds in paper documents, and the fonts, colour, image sizing, and other key characteristics. This kind of technology is also being used to source photographs of consumers from social media sites or other websites and to reformat and resize them to create fake ID documents.

Generative AI can be used at scale to create fake online footprints for consumers and businesses – including fake social media accounts and websites – making it seem as though applications are coming from genuine customers. This makes it more difficult to defeat fraud prevention methods based on ‘web crawling’, leading to an increased risk of fraud and related financial losses.

As well as using generative AI to create identities and falsify documents, systems can be used to create mass scams, drafting realistic emails, letters and texts to inform unsuspecting consumers of supposed unusual activity on their accounts. What’s more, using sophisticated tracking and monitoring systems, fraudsters are able to use AI to further refine and target communications to those most susceptible to take action and fall into the fraudster’s trap.

One of the biggest concerns about AI-powered fraud techniques is their ability to ‘learn’. They do this by analysing the causes of each failed fraud attempt and making adjustments based on a new understanding of the strengths and weaknesses of an institution’s fraud detection systems. This makes AI-powered fraud attacks increasingly difficult to address and defend-against over time.

Using AI to fight AI-powered fraud

To detect and prevent fraud involving identities or documents created using generative AI, institutions need to deploy AI solutions themselves. This approach gives organisations the edge in terms of analysing data from fraud incidents and attacks on a huge scale, with the ability to ‘learn’ about new fraud techniques and, crucially, how to defend against them.

Importantly, this kind of AI-driven analysis also needs to take a multi-layered approach that considers a number of factors, from close examination of consumers IDs and identity documents to the devices they are using to make applications and their online footprints.

1. Cross referencing images and consumer PII with other trusted data sources for verification purposes
   By using AI algorithms to cross-reference information from consumer applications with trusted data such as credit bureau data, banking data, the electoral roll and UK Government databases, it is possible to identify anomalies and to ensure that fraud is detected at the earliest stage possible.
2. Using device biometrics techniques and technologies
   There are countless cases of thousands of fraudulent applications being made from a single device and single IP address. This kind of device data, processed by AI tools, can help organisations quickly identify fraud and stop attacks immediately.
3. Reviewing and validating consumers’ online accounts and footprints
   There are typically significant differences between genuine customer online accounts and activity, and the kinds of fake digital footprints that can be generated by ML and AI algorithms. More advanced defence solutions can distinguish between genuine consumer online activity and fictitious accounts, helping organisations to identify and stop potential fraud.
4. Conducting perpetual Know Your Customer (KYC) checks
   Bearing in mind that many fraud attacks start small and make regular payments to increase credit limits before defrauding institutions, KYC checks need to be conducted regularly, on an ongoing basis, to mitigate fraud risks and associated losses. Automated checks using AI algorithms can help to detect these so-called ‘escalation attacks’.
5. Deploying the very latest AI-powered technologies and toolsets
   With fraudsters using ever more sophisticated generative AI tools to create synthetic identities and documentation, organisations also need access to the most advanced, automated fraud-detection and prevention technologies and AI models to stay one step ahead – requiring constant analysis of fraudsters’ use of AI and ongoing investments in the latest AI algorithms and solutions.

How can we help?

Experian offers the multi-layered solutions organisations and consumers need to minimise the negative impacts of fraud involving generative AI.

Specifically, we use data from multiple trusted sources to enrich our own extensive credit bureau data on the UK population, and to verify veracity of consumer identities online. Additionally, we can use device fingerprints and web-crawling technology to identify potential incidents of fraud, reducing the risk of multiple fraudulent applications being made from a single device or location and ensuring that customers have a genuine online presence that indicates that they are who they say they are.

This multi-layered approach to fraud prevention and detection means that institutions can identify fake and synthetic identities and applications and minimise fraud risks and losses.

As well as helping to detect fraud, our machine learned solutions help genuine customers – and especially customers in difficult-to-authenticate categories such as young people, people who are new to the UK, or customers who do not own their own home – to access the financial products and services they need. This is thanks to our rich data on consumers across the UK and our advanced analytics capabilities, which allow us to differentiate genuine customers from fake or synthetic IDs quickly and at scale.