AI in Banks

Sony Bank and Fujitsu applied Anthropic's Claude on Amazon Bedrock, with Claude Code and MCP-connected agents, to the development of Sony Bank's production core banking system from September 2025. By July 2026 the development period from basic design to integration testing was 30% shorter and man-hours 40% lower, with humans keeping final judgement and quality assurance.

Starling Bank moved its data warehouse to BigQuery and its machine learning to Vertex AI, then built Gemini-based tools: Scam Intelligence checks marketplace ads for scam signs and has raised payment cancellations by 300%, call-transcript summaries for staff save 8,000 hours a month, and a Gemini help centre cut customer service referrals by 50%.

Chime Financial deployed Anthropic Claude via Amazon Bedrock to automate customer call summarization, cutting average agent handling time by 18 seconds per call and saving 250,000+ hours annually (~$700K in efficiency gains), with NPS rising 5 points after the February 2024 production launch.

AI-driven AML reduces false alerts and speeds investigations when banks align data, models and workflows, with validation and human oversight. This editorial from Asset Management News surveys how banks reshape compliance programs around generative AI, focused on cutting false positives while keeping regulator-grade traceability. The piece draws on European banking deployments and notes both the upside (alert time cut, investigation speed) and the cost (case-handler training, model governance). Markets including Germany, the UK, and Switzerland are highlighted as adopting AI-driven alert scoring faster than the US retail-banking segment.

Bank of Kyoto, working with NTT Data, trained its own model in DataRobot on past suspicious transactions using transaction and customer data on its SIC data platform, and built a dashboard that puts the model's predictions and the data investigators need in one view. In validation from December 2025 it found cases earlier methods missed and cut follow-up work by about 30%.

JPMorgan Kinexys and BNY jointly explored federated learning for cross-institution anomaly detection in a permissioned PoC using synthetic data, with the aggregated model converging by the fourth aggregation round and detecting anomalies beyond either participant's local model.

First National Bank of Omaha piloted Pindrop's Fraud Assist, a GenAI add-on to Pindrop Protect, from August to December 2025. Pindrop reports 35-40% faster investigations, 42% more cases handled by a 4-5-person team without added staff, and 50% higher analyst accuracy.

Rabobank's Digital Platform engineering team has been experimenting with AI agents and multi-agent architectures on Azure AI Foundry and the Microsoft Agent Framework to evaluate production readiness. The team documents recurring challenges — supervisor-agent dilution, tool-call failures, non-deterministic orchestration — and shares lessons on prompt design, model selection and continuous evaluation.