Swift, in collaboration with 13 international banks and technology partners like Google Cloud, conducted experiments demonstrating how AI and privacy-enhancing technologies (PETs) can significantly reduce cross-border payment fraud by enabling secure, real-time data sharing.

These experiments involved using PETs to allow global financial institutions to share fraud insights securely across borders, which helps verify suspicious accounts instantly and accelerates fraud detection from hours or days to minutes.

Industry representatives, including BNY, emphasized that these technological advancements promote security and elevate standards within the cross-border payments ecosystem, ultimately reducing costs associated with fraud.

ANZ's head of technology highlighted that federated learning, a key component of these innovations, strengthens defenses against rising global fraud and scams.

Rachel Levi, Swift’s Head of AI, stressed that industry-wide cooperation in fraud defense is crucial, as it enhances security, speeds up response times, and could save billions lost annually to financial crime.

Swift currently has over 50 AI use cases, including a Payments Controls Service for smaller institutions, as part of its broader effort to combat the estimated $485 billion lost annually to financial crime.

Major banks like ANZ, BNY, and Intesa Sanpaolo, along with Google Cloud, participated in these experiments, demonstrating strong industry support for advancing cross-border fraud prevention technologies.

The AI-based methods tested can detect complex financial crime networks faster, reducing fraud response times from hours or days to minutes, which could save the industry billions annually.

Intesa Sanpaolo highlighted that fraud increases friction and costs in cross-border payments, and these technological innovations aim to improve security and reliability.

One of the experiments enabled real-time verification of suspicious accounts, allowing for quicker identification and prevention of fraudulent transactions before they occur.

Following successful trials, Swift plans to expand participation and incorporate real transaction data in future testing to evaluate the impact of these technologies in real-world scenarios.

A combined use of PETs and federated learning trained AI models on synthetic data from 10 million transactions, showing twice the effectiveness in fraud detection compared to models trained on individual institutions' data.