JPMorgan Chase and Bank of America are using synthetic transaction data to train fraud detection models without exposing customer records. The shift signals a pragmatic path to faster model iteration and safer collaboration under tightening privacy and compliance expectations.
Major banks adopt synthetic transaction data to train fraud models
JPMorgan Chase and Bank of America are partnering with synthetic data providers to improve fraud detection model training while protecting customer privacy. Instead of relying solely on production transaction logs, the banks are adopting synthetic transaction datasets to boost AI performance and to enable development and testing workflows that reduce exposure of sensitive customer information.
The move reflects a broader pattern in financial services: teams want more data to train and validate fraud models, but direct use of real customer records increases privacy risk and can slow experimentation due to access controls and regulatory scrutiny. Synthetic datasets are positioned as a way to expand training coverage (including edge cases) while keeping development environments and test pipelines away from regulated data.
- Faster iteration with fewer approvals: Synthetic datasets can reduce dependence on tightly controlled production data, helping fraud and ML teams test features, model changes, and monitoring workflows without repeatedly requesting access to sensitive records.
- Lower privacy exposure in model development: Using synthetic transactions can support privacy-by-design practices by limiting how often real customer data is copied into dev/test environments.
- Safer cross-institution pattern sharing: Synthetic data can make it more feasible to share fraud pattern signals across institutions with reduced privacy risk, potentially improving detection of emerging schemes.
- Compliance alignment remains a gating factor: Even with synthetic data, banks still need governance over how datasets are generated, validated, and used to ensure they meet internal risk standards and external regulatory expectations.