Industry guides
Private AI for US banks and credit unions: GLBA, NYDFS Part 500 and model risk on a governed data platform
US financial institutions can adopt AI agents if access to nonpublic personal information is controlled and evidenced. GLBA's Safeguards Rule requires access controls and safeguards, NYDFS Part 500 brings AI systems into covered entities' cybersecurity programmes, and model risk guidance expects inventories, validation and monitoring. A governed lakehouse with row filters, masks, lineage, logged agent actions and a model registry produces that evidence.
Published . Last updated
On this page
US banks and credit unions are not prohibited from using generative AI. They are expected to govern it like any other system that touches nonpublic personal information (NPI). This guide maps the main US expectations to the controls a data and AI platform must provide, and shows how DataNature meets them in a US deployment.
What do US regulators expect from AI that touches customer data?
| Framework | What it expects | Platform control |
|---|---|---|
| GLBA Safeguards Rule | Safeguards and access controls for customer information | Row filters, column masks, least-privilege groups, encryption |
| NYDFS Part 500 | Covered entities include AI systems in their cybersecurity programme: asset inventory, access privileges, audit trails, third-party risk, incident response | Inventory of twins and models, scoped access, audit logs, AI gateway as a governed dependency |
| Model risk management guidance (SR 11-7) | Model inventory, development standards, validation, ongoing monitoring | Model registry, reproducible snapshots, lineage, decisions written back for monitoring |
| State privacy laws such as CCPA | Rights and transparency for consumers' personal information | Catalog and lineage to find personal data; masking and deletion |
NYDFS confirmed in its October 2024 industry letter that existing Part 500 obligations apply to AI, and its May 2026 publications on frontier AI risk map to the same obligations rather than creating new ones.
Why is the data layer where AI compliance is won or lost?
Because an agent inherits whatever access its data path allows. If the agent's connector can read every table, prompt instructions will not reliably stop it. The controls that hold are enforced below the model:
- Row filters and masks applied by the SQL engines per user or group.
- Twin scopes that limit an agent to selected catalogs, schemas and tables.
- SQL allowlists and document partitions enforced inside each tool call by Voir's Agent Meta-Harness.
- Human approval for any action that changes a system, with every action logged.
Can a US institution run this privately?
Yes. DataNature runs in US regions of your cloud account (for example IBM Cloud, AWS, Azure or Google Cloud), in a dedicated environment, on-premises or air-gapped. Models are reached through an AI gateway pointed at your approved endpoint, which can be a self-hosted model on your own GPUs. US customers can buy and deploy DataNature directly; see DataNature in the United States.
Which use cases do US banks and credit unions start with?
- Member and customer 360 with next-best-product lists for branch and contact-centre staff.
- Real-time fraud and account takeover detection on the StreamHouse hot path.
- AML investigations with network analysis on the knowledge graph and cited case narratives.
- Loan and member document automation with DocuContext.
- Board and regulatory reporting on certified cubes with lineage.
Frequently asked questions
Is DataNature compliant with GLBA and CCPA?
DataNature provides the safeguards these frameworks expect, such as access controls, encryption, masking, lineage and audit. Compliance is assessed for each institution's programme, and DataNature is designed to support it.
Does DataNature use our data to train models?
No. Customer data is not used for model training or shared with other customers.
How does this help with examinations?
Examiners ask who can access NPI, how models are inventoried and monitored, and what AI did. Atlas audit logs, the model registry, lineage and Voir action logs answer those questions with evidence.
Sources
- Kiteworks: NYDFS Part 500 and AI agent governance
- Sidley: NYDFS coordinated guidance on frontier AI cybersecurity risks
- DataNature documentation: security overview
- DataNature: United States region
How does DataNature Atlas keep this secure and compliant?
One set of controls supports banking, privacy and AI rules at the same time. DataNature Atlas is the platform's governance module:
- Knows your data: a catalog with automated PII detection and classification across Bronze, Silver and Gold.
- Controls access: role-based grants plus attribute-based policies, row filters and column masks (null, redact, partial, hash, year only).
- Governs AI: Voir agents are scoped principals; masks, row filters and PII guardrails apply to every tool call and answer.
- Proves it: lineage from source to metric to decision, and an audit log of who or what accessed which data.
- Stays local: Atlas runs inside your deployment, so policies, lineage and logs never leave your jurisdiction.
These controls support GDPR, PIPEDA, Quebec Law 25, CCPA, HIPAA and sector rules such as OSFI E-23. Read DataNature Atlas in depth or see security and compliance.
Platform components in this article
Learn more about each DataNature product and module mentioned here.
Products
See it on your own data.
Book a demoDemos and quotes are handled by Dlytica, the company behind DataNature.