Starling Bank, an innovative player in the financial sector, has recently unveiled its trailblazing tool titled “Scam Intelligence”. This pioneering tool embraces the power of generative artificial intelligence (AI) to enable customers to upload texts and images for the purpose of identifying potential signs of common payment scams. According to the bank’s claims, this is the first tool of its kind in the United Kingdom.
Scam Intelligence analyses uploaded content using Google’s advanced Gemini models and provides personalized risk guidance for the user. It can identify warning signs such as suspiciously low prices, attempts to rush the customer into action, or payment details that do not align with those of a legitimate vendor. The primary goal of this feature is to protect customers from authorized push payment (APP) fraud, which is a significant issue in the UK.
Understanding Authorized Push Payment Fraud
Authorized push payment (APP) fraud is a type of scam where fraudsters manipulate customers into transferring money directly into their bank account. Unlike unauthorized fraud where the criminals gain unauthorized access to accounts or steal checks, in APP fraud, the customer authorizes the payment, often under false pretenses. This can take the form of a romance scam where the victim is manipulated into sending money to help their ‘partner’ in an emergency.
As per the trade group for the UK’s financial services sector, UK Finance, APP fraud resulted in losses of £450 million (approximately $590 million) in the year 2024 alone. This staggering figure underscores the importance of proactive measures like Scam Intelligence to combat such scams.
Leveraging the Power of Generative AI
Starling Bank’s Scam Intelligence tool harnesses the prowess of Google’s Gemini models, which were developed as a response to OpenAI’s ChatGPT. The tool operates on the Google Cloud platform, utilizing these models to comprehend the context behind the uploaded images and texts. Starling Bank’s proprietary system then provides the final risk assessment to the customer.
The UK’s Minister for Fraud, David Hanson, has praised the tool, describing it as an excellent example of leveraging AI in the fight against fraud. Using retrieval-augmented generation (RAG), a large language model can reference a knowledge base before generating a response. This could prove invaluable in identifying scams, as the knowledge base could include documents outlining the tactics used by criminals in various scams.
UK Regulatory Pressure Versus US Liability
The UK has stringent regulations that require banks to reimburse most APP fraud victims. The Payment Systems Regulator (PSR) mandates that regulated companies reimburse victims, with the cost of reimbursement equally divided between the sending and receiving bank.
Contrastingly, in the US, the Electronic Fund Transfer Act does not obligate banks to reimburse consumers for losses incurred due to APP fraud. While some banks voluntarily agree to reimburse victims of specific imposter scams, they are generally not liable for such losses.
Comparing Scam Intelligence with ThreatMark’s ScamFlag
Starling Bank’s initiative follows the launch of a similar tool, ScamFlag, by fraud prevention company ThreatMark. Like Scam Intelligence, ScamFlag uses AI to detect scams and can be used on multiple digital channels.
ThreatMark trained its generative model on scam samples, enabling it to analyze pictures, extract visible text, and check identified links or bank account numbers against its database. The company claims an impressive 99% accuracy rate in detecting scams and offers ScamFlag as a software development kit (SDK) for integration with existing mobile applications.
In conclusion, the introduction of AI-powered tools like Scam Intelligence and ScamFlag marks a significant step forward in the fight against APP fraud. These tools not only empower customers to identify potential scams but also align with the regulatory requirements of banks, especially in regions where they bear the liability for such frauds.
Source: Here
