Half a century after transforming financial messaging, Swift is testing whether privacy-enhancing AI can underpin a global shield against cross-border fraud.
When Swift was founded in 1973, its mission was to create a secure, standardised messaging system which could replace the patchwork of telexes and phone calls then used by banks.
Today, more than 11,500 institutions across 200 countries rely on its infrastructure, making it the backbone of global payments.
Yet as Swift marks its 50th year, the challenge has shifted from messaging reliability to criminal resilience. Fraud has become one of the costliest risks in finance, with an estimated $485bn lost globally in 2023, according to the firm. Sophisticated scams, mule networks and data breaches increasingly cross borders faster than banks can respond.
Against this backdrop, Swift has unveiled the results of a year-long set of experiments that could point to a new frontier in fraud prevention: AI-powered collaboration across borders. Working with 13 international banks including ANZ, BNY Mellon and Intesa Sanpaolo, and technology partners such as Google Cloud, Swift tested the use of privacy-enhancing technologies (PETs) and federated learning models on ten million synthetic transactions.
The results, announced on September 15, showed that by allowing banks to share insights without exposing underlying customer data, the AI models were twice as effective at spotting fraudulent activity as those trained on a single institution’s dataset. I
n one use case, participating banks could verify suspicious accounts in real time, potentially stopping funds before they moved through complex criminal networks.
Rachel Levi, Head of AI at Swift, framed the results as evidence of the cooperative’s convening power. “These experiments demonstrate the convening power of Swift as a trusted cooperative at the heart of global finance,” she said. “A united, industry-wide fraud defence will always be stronger than one put up by a single institution acting alone.”
A history of secure collaboration
Swift has long positioned itself as more than a messaging provider, often stepping into roles that require industry-wide coordination. After the 2016 cyberattack on Bangladesh Bank, which exposed vulnerabilities in banks’ endpoint security, Swift launched its Customer Security Programme, mandating stronger defences across its network.
More recently, it has expanded into fraud and compliance tools. Its Payments Controls Service, rolled out in 2019 and enhanced with AI earlier this year, enables institutions to screen transactions for anomalies before execution. According to Swift, hundreds of small and medium-sized banks now rely on it as their frontline fraud filter.
What differentiates the latest experiments, however, is the ambition to link institutions together rather than improve individual resilience. By using PETs, which allow data to be analysed without leaving the owner’s system, and federated learning, which “visits” each institution to train models locally, Swift argues that global collaboration on fraud no longer requires compromising privacy.
Industry voices signal appetite
Participants in the trials were quick to highlight the significance of a cooperative model. “The rise in fraud and scams is a global issue impacting all financial institutions,” said David Buckthought, Head of Technology – Payment Services and Digital Assets at ANZ. “This will provide banks with a stronger defence against fraudulent activity.”
BNY Mellon’s Isabel Schmidt described the trials as proof that “competitive organisations can come together behind a greater good, while driving standards in security and enhancing the experience of all stakeholders.”
For Intesa Sanpaolo, one of Europe’s largest retail banks, the trials demonstrated how “a synergistic approach supported by the latest technologies” could reduce friction and costs across the payment ecosystem, according to Enrico Canna, head of its Anti-Fraud and Customer Protection Centre.
The wider policy context
Regulators in the UK, EU and US have all pressed financial institutions to strengthen their fraud prevention frameworks in response to rising scam losses. In the UK, the Payment Systems Regulator (PSR) will in 2025 enforce mandatory reimbursement for authorised push payment fraud, shifting costs back to banks and increasing the incentive to block suspicious transfers earlier.
At the global level, the Financial Stability Board (FSB) has repeatedly warned fragmented approaches to cybercrime and fraud are undermining financial stability. While information sharing has long been a policy goal, concerns about data protection and liability have hampered progress. Swift’s trials suggest that PETs could offer a compliant pathway.
Swift’s AI trajectory
The fraud initiative is part of a broader exploration of artificial intelligence by Swift. The cooperative says it now has more than 50 AI use cases across proof of concept, pilots and live services, spanning areas from anomaly detection to operational efficiency.
Its long-term challenge, however, mirrors that of the wider industry: moving from pilots to scale.
For Swift, the project reflects both continuity and reinvention. Just as its founders built a secure backbone for banking messages in the 1970s, the cooperative is now attempting to convene the industry around fraud resilience.
Whether it can deliver on the promise of “fraud stopped in minutes, not days,” as Levi put it, will depend on more than algorithms. It will require banks, regulators and technology providers to accept that fighting financial crime is one area where cooperation must trump competition.
