How Swift's AI experiment works
Swift, a cooperative at the heart of the global financial messaging, has uncovered the results of a groundbreaking experiment designed to combat fraud with international payments. The initiative involved 13 financial institutions working with to test the use of artificial intelligence and privacy-enhancing technologies (pets) in fraud detection. By enabling secure sharing of fraud-related insights across borders, these experiments aim to reduce losses and accelerate the speed at which fraudulent activities can be intercepted.
One of the major use cases involved real-time checking of intelligence on suspicious accounts. Pets allow participants to safely share insights and identify potential financial crime networks before fraudulent transactions are carried out. This represents the transition from siloed intelligence to cross-border cooperative fraud defense.
In another test, federal learning was applied to a synthetic dataset representing 10 million artificial transactions. The models were trained locally at each institution without revealing customer data. Compared to the single-center model, this collaborative approach is twice as effective in detecting known fraud patterns, suggesting a powerful application of large-scale cross-border fraud protection.
remove
Federated Learning and Pets doubled the effectiveness of fraud detection and proved the value of secure, cross-border AI collaboration in financial services.
Why cross-border collaboration is important
Cross-border payment fraud remains an urgent issue, with financial crimes costing an industry that is estimated at $48.5 billion in 2023. Swift's experiment highlights that fragmented responses by individual banks cannot match the size of fraud networks operating across jurisdictions. A unified approach that leverages AI and privacy tools can significantly reduce detection times and industry-wide losses.
Rachel Levi, head of AI at Swift, emphasized the power of collective action. “These experiments demonstrate Swift's convened power as a trusted cooperative at the heart of global finance. The unified industry-wide fraud defense is stronger than that always posted by a single institution acting alone.” By allowing institutions to exchange intelligence in minutes rather than days, Swift paves the way for a faster, safer global payment system.
Swift is actively investigating AI applications, with over 50 projects piloting or producing, including the Ai-Enhanced Payments Controls Service launched earlier this year. The service already helps small and medium-sized banks flag suspicious transactions in real time, highlighting Swift's ongoing push for AI.
remove
What participants and partners say
Several major institutions and technology providers participated in the experiment, including ANZ, BNY, Intesa Sanpaolo and Google Cloud. Their involvement underscores the broad industry's desire for joint solutions that protect consumers and institutions alike. Feedback from participants suggests optimism about AI-driven fraud defense becoming a core component of future payment infrastructure.
David Buckthought of Anz said: “The rise in fraud and fraud is a global issue affecting all financial institutions. ANZ is excited to be involved in the response across the industry, proving its use of federal learning to increase detection capabilities. His remarks underscore the importance of competitive interagency collaboration to create systematic resilience.
Reflecting the sentiment, BNY's Isabel Schmidt said: “Security is paramount in cross-border payments. Using the latest technology, the group has achieved results showing how these tools can be used to lift the entire ecosystem and demonstrate the value of a rapid cooperative. Meanwhile, Enrico Canna of Intesa Sanpaolo added: “Trans-border payment fraud increases ecosystem friction and causes significant costs at the industry level. Intesa Sanpaolo is working together in these preliminary experiments led by Swift to demonstrate the positive impact of a synergistic approach supported by modern technology that is safer and more reliable.”
remove
Global Banks supports Swift's AI-driven approach, citing a cooperative model that offers stronger security and benefits the entire payments ecosystem.
Next up on Swift's scam program is
Following the success of the initial exam, Swift plans to expand participation and move towards actual testing using live transaction data. The second phase examines how pet-federation learning models work in operational environments where speed and accuracy directly affect financial outcomes. If successful, the technology will allow banks around the world to block fraudulent payments in real time.
Industry observers see this as a major step forward in modernizing fraud defense. By using guardrails to responsibly use AI for privacy and compliance, financial institutions can better protect their customers without compromising sensitive data. Such systems can also alleviate regulatory concerns by providing more consistent fraud detection criteria across jurisdictions.
The long-term vision is a cooperative fraud shield built on shared intelligence, real-time data processing, and AI models that can evolve with new threats. In the case of global finance, this could mark the beginning of a new era where fraud detection is not the responsibility of one agency, but a shared mission across the ecosystem.
remove
The next step in Swift is real-world testing, aiming to turn AI experiments into a cross-border fraud defense system for the global banking industry.
