Technology & AI
AI deployment in core banking operations, model governance, payments infrastructure, systemic risk from algorithmic concentration, and the technology decisions that are quietly reshaping global finance.
-
12 Mar 2026JPMorgan Chase · Credit
JPMorgan AI System Now Processes 62% of Consumer Loan Applications Without Human Review
JPMorgan Chase disclosed in a regulatory filing that its AI-assisted credit underwriting system now processes 62% of all consumer loan applications without human review. The system, deployed across the retail bank since January, has reduced average decision time from 4.3 days to 11 minutes. Default rates on AI-approved loans are tracking 18 basis points below the human-reviewed cohort after two months.
Continue reading -
11 Mar 2026SWIFT · Cross-Border Payments
SWIFT Launches 23-Bank AI Routing Consortium for Cross-Border Payments
SWIFT announced a consortium of 23 banks will begin live testing of its AI-powered cross-border payment routing system in April. The system uses a transformer model trained on three years of settlement data to predict optimal routing paths, with the aim of reducing failed transactions by 40%. Deutsche Bank, Standard Chartered, and BNY are among the first participants.
Continue reading -
10 Mar 2026Bank of England · Systemic Risk
Bank of England Warns of ‘Correlated Failure Modes’ From AI Model Concentration
The Bank of England published a discussion paper on "algorithmic systemic risk," warning that the concentration of financial decision-making in a small number of foundation models could create correlated failure modes that existing stress-testing frameworks cannot capture. The paper invites responses by 30 June and signals that model-level regulation is under active consideration.
Continue reading -
9 Mar 2026HSBC · Trade Finance
HSBC Trade Reconciliation AI Reduces Exception-Handling Labour by 40%, Equivalent to 1,200 Roles
HSBC has disclosed that a transformer-based reconciliation model deployed across its global trade finance operations has reduced the labour required for exception handling by 40%, equivalent to approximately 1,200 full-time roles. The system identifies mismatches in trade documentation with 99.2% precision, against 94.7% for the previous rules-based system.
Continue reading