DataPacts

Banks and Financial Firms Licensing Data to AI: GLBA and Secrecy Rules

Checked

At a glance

A bank or financial firm can license some of its records to AI developers, but customer financial privacy law, suspicious activity report confidentiality, confidential supervisory information and, for Swiss and EU banks, statutory secrecy and operational rules decide which records can leave.

Independent reference. Every buyer is covered on the same terms, every figure carries its source and check date, and no buyer pays to be listed or placed higher.

Request an independent introduction

Tell us what data your company holds. We reply with the buyers that fit and what their terms leave out. Your details reach a buyer only after you agree to a named introduction.

IndependentYour details stay with us until you agreeNot selling is a valid outcome

HUMXN names loan files, underwriting decisions, transactions and fraud reviews and Handshake AI names underwriting workflows on their own pages, checked 2026-11-17, and none of the four program pages compared below states a price for bank data.

The rules below are stated as in force on 2026-11-17, and every buyer statement is a dated claim from its own page. This page is information, not legal advice, and it sits within AI training data licensing, where bank records are one sector among many.

How we checked this page: each citation was compared with its primary source on the date shown, and the method is set out under how we source, date and check every figure.

What Banks and Financial Firms Hold That AI Buyers Name

Four program pages name or exclude records that banks and financial firms hold, and none of them is a bank-specific price list: Handshake AI, HUMXN, Mercor's Payments & Fintech listing and micro1's exclusion clause. The table gives their own words, as of 2026-11-17, in alphabetical order.

Program Words on its own page naming bank-held records Condition or exclusion stated Checked Source
Handshake AI Data Partnerships "Finance & insurance: Financial models, process documentation, underwriting workflows" Size conditions differ across its pages: no minimum on the Data Partnerships page, 20+ full-time employees and 3+ years in operation on the license and referral pages 2026-11-17 (size conditions 2026-11-09) joinhandshake.com/ai/data-partnerships
HUMXN "Banking & finance: Loan files, underwriting decisions, transactions, fraud reviews" No exclusion stated on the page read 2026-11-17 humxn.io/sell-company-data
Mercor, Payments & Fintech listing "transaction, order, inventory or shipment tables" from a "payment processor, merchant acquirer, POS, e-commerce platform, logistics or market-data company" 3 to 5+ years of history and a consistent schema; Mercor states USD 80,000 to 900,000 for the listing as of 2026-11-17 2026-11-17 api.data.mercor.com listing feed
micro1 Data License Master Terms s.5.7(e): the data will not contain "payment card data or financial account data, including bank account, debit card or credit card numbers or associated security or access codes" Terms "Last updated: November 12, 2026" 2026-11-17 micro1.ai/legal/data-license-master-terms

Mercor's listing asks for payment processors, merchant acquirers and similar firms; in the feed read on 2026-11-17 its Investment Firm listing names boutique investment banks and private credit, and no listing names a credit union or consumer loan records. A "not stated" cell never means cleared. Terms across buyers sit under company data partnership programs.

The Rules Between a Bank Record and a Buyer

Each bank record family meets a different rule before a buyer's license, and the table pairs the record buyers name with the rule that stands in front of it. "None named" means no program page read names that record, not that the rule is moot.

Bank record family Named by Rule that stands in front of it Citation (in force as of 2026-11-17) Section below Source
Consumer loan, deposit and card records HUMXN (loan files, transactions) Notice and opt-out before disclosure to a nonaffiliated third party 15 U.S.C. 6802(b)(1); 12 CFR 1016.10 Consumer Customers ecfr.gov; govinfo.gov
Business and commercial customer files HUMXN (loan files) Outside Regulation P; contract and other law apply 12 CFR 1016.1(b)(1) Consumer Customers ecfr.gov
Underwriting decisions and workflows Handshake AI, HUMXN Depends on whether applicant data remains; blind data without identifiers is not personally identifiable financial information 12 CFR 1016.3(p)(1), (q)(2)(ii)(B) Consumer Customers ecfr.gov
Transaction tables HUMXN, Mercor Payments & Fintech Nonpublic personal information where derived from consumer information; micro1 excludes account and card numbers; bulk threshold of more than 10,000 US persons 12 CFR 1016.3(p)(1); micro1 s.5.7(e); 28 CFR 202.205 Consumer Customers ecfr.gov; micro1.ai
Fraud reviews and SAR-linked files HUMXN Confidential if they would reveal a SAR 31 CFR 1020.320(e)(1)(i); 31 U.S.C. 5318(g)(2) Suspicious Activity Report Files ecfr.gov; govinfo.gov
Examination reports, supervisory letters and documents quoting them None named Confidential supervisory information 12 CFR 261.20(a), 261.2(b)(1), 4.37(b)(1), 309.6(a) Confidential Supervisory Information ecfr.gov
Swiss bank client data None named Bank secrecy, Banking Act Art. 47 (SR 952.0), version of 2026-10-01 BankG Art. 47 Swiss Bank Secrecy fedlex.admin.ch
EU bank customer data and delivery tooling None named GDPR purpose compatibility; DORA contract terms for ICT services GDPR Art. 6(4); Regulation (EU) 2022/2554 Arts. 2, 3(21), 28, 30 EU Banks eur-lex.europa.eu

Consumer Customers and Business Customers: Where Regulation P Stops

Regulation P protects consumers only: 12 CFR 1016.1(b)(1) says the part does not apply to information about companies or about individuals who obtain financial products or services for business, commercial or agricultural purposes (eCFR as of 2026-11-01).

A consumer loan, deposit or card record is nonpublic personal information, as is any list or grouping of consumers derived from it (12 CFR 1016.3(p)(1)), and disclosure to a nonaffiliated third party needs notice and a chance to opt out first (15 U.S.C. 6802(b)(1)).

The notice steps, exceptions and reuse limits are set out under GLBA and financial customer data.

A commercial loan file falls outside Regulation P but not outside every duty, because the borrower contract, counterparties' confidential terms and the GDPR for EU individuals can still apply.

Suspicious Activity Report Files and Fraud Reviews

A bank may not disclose a suspicious activity report or any information that would reveal the existence of one (31 CFR 1020.320(e)(1)(i), eCFR text as of 2026-11-01).

The rule names the bank and its directors, officers, employees and agents, and a bank asked to disclose declines, cites the section and 31 U.S.C. 5318(g)(2)(A)(i), and notifies FinCEN.

HUMXN names "fraud reviews" among bank records on its company data page, checked 2026-11-17. A review file that records a filing or a decision to file would reveal a report's existence, a consequence of the quoted text.

Finance records, including this limit, are summarized under financial records for AI training, where Regulation S-P is also covered.

Confidential Supervisory Information Inside Bank Documents

Documents that quote or reveal a supervisor's examination findings are confidential supervisory information, and the Federal Reserve, the OCC and the FDIC each bar disclosure without their approval.

Under 12 CFR 261.2(b)(1), it includes reports of examination and supervisory correspondence, and any portion of a document held by anyone, including the institution, that would reveal it is itself confidential supervisory information.

Under 12 CFR 261.20(a), no person possessing it may disclose it without the prior written permission of the Board's General Counsel (eCFR as of 2026-11-01).

For national banks and Federal savings associations, 12 CFR 4.37(b)(1) bars disclosure of OCC non-public information without OCC approval, except after seeking it from the OCC and as a federal court orders, and (b)(1)(ii) refers to penalties under 18 U.S.C. 641.

Section 4.37(b)(2) allows disclosure to persons officially connected with the bank and to a consultant under written contract with the bank, and a data licensee is not among the persons named in the text read. For the FDIC, 12 CFR 309.6(a) bars disclosure of exempt records without written authorization from the responsible Division Director.

A document prepared for the institution's own business purposes, held by it and containing no confidential supervisory information, is not confidential supervisory information (12 CFR 261.2(b)(2)(i)). Policies, risk reports and model reviews are therefore screened for quoted findings before they enter a dataset. None of the program pages read mentions this limit.

Swiss Bank Secrecy Under Article 47 of the Banking Act

Article 47 of the Swiss Banking Act (SR 952.0) makes it a crime for a bank's organ, employee, agent or liquidator to intentionally disclose a secret entrusted to them in that capacity or perceived in it, punishable by a custodial sentence of up to three years or a monetary penalty.

In the German text of the Federal Act on Banks and Savings Banks, version of October 1, 2026, paragraph 1bis sets up to five years or a monetary penalty where a pecuniary advantage is obtained through 1(a) or (c).

Negligent disclosure carries a fine of up to CHF 250,000 in that version (paragraph 2), and liability continues after the end of the service relationship (paragraph 4).

Paragraph 5 reserves federal and cantonal rules on testimony and on information duties to authorities. The article's text does not mention anonymized or aggregated data, so a Swiss bank checks it before any license. The wording here paraphrases the German text.

The data protection act, copyright and employee rules are covered under Swiss rules for licensing data.

EU Banks: Purpose Limits and DORA Contracts

An EU bank needs an Article 6 legal basis for every disclosure of customer personal data to an AI developer and, for data collected for another purpose, a compatibility test under Article 6(4) of the GDPR.

The test weighs the link between the purposes, the context, the nature of the data, the consequences and the safeguards.

Regulation (EU) 2022/2554 (DORA) applies from 2025-01-17 (Art. 64) and covers credit institutions (Art. 2(1)(a)), payment institutions (b) and investment firms (e). It regulates the ICT services a financial entity uses, meaning digital and data services provided through ICT systems on an ongoing basis (Art. 3(21), Art. 28(1)(a)).

Contracts for those services must state the locations of processing and storage (Art. 30(2)(b)), data protection provisions (30(2)(c)) and access, recovery and return of data on provider insolvency or termination (30(2)(d)), which matters where an ICT provider extracts, stores or transmits a dataset.

The Regulation's text does not use the words artificial intelligence, machine learning or training data, and it does not address licensing data to a developer. Compatibility testing is set out under GDPR purpose compatibility.

Benefits, Risks, Downsides and Alternatives for a Bank

Licensing bank records can put history the bank already keeps to use, but customer privacy law, secrecy statutes and supervisory confidentiality remove many of the records buyers ask for, and a trained model cannot be untrained.

Benefits

  • HUMXN and Handshake AI name process and decision records, such as underwriting decisions and workflows, checked 2026-11-17.
  • Procedure and model documents with no customer data raise no notice and opt-out question.
  • Mercor states USD 80,000 to 900,000 for its Payments & Fintech listing as of 2026-11-17, a claim for payment and commerce firms.

Risks

  • Disclosing nonpublic personal information without notice and an opt-out chance (15 U.S.C. 6802(b)(1)).
  • Revealing the existence of a suspicious activity report (31 CFR 1020.320(e)(1)(i)).
  • Disclosing confidential supervisory information, with 18 U.S.C. 641 referenced in 12 CFR 4.37(b)(1)(ii).
  • Criminal exposure for staff under Article 47 of the Swiss Banking Act.

Downsides

  • micro1's s.5.7(e) excludes account and card data, which removes a record family from its set, checked 2026-11-17.
  • Labelbox Prism states exclusivity by default and that trained models "cannot be unlearned", checked 2026-11-09.
  • Aba Compute states that deletion does not reverse completed training, checked 2026-11-09.

Alternatives

  • License procedures and model documents that hold no customer data.
  • License a de-identified subset; Replay states it takes only de-identified data, checked 2026-11-09.
  • Choose non-exclusive terms; Pangea.ai states it licenses non-exclusively and the company keeps ownership, checked 2026-11-09.
  • Keep the records: not licensing at all keeps every right and is a valid outcome.

Where Bank Licensing Meets Insurers, Investment Firms and the Seller Path

Bank licensing is one industry among several that hold customer financial records, and the neighboring pages cover the records and rules this page leaves out. The index of data licensing by industry compares sectors.

Two pages cover data and law that cut across banks:

The seller path for any bank record runs through five steps:

  1. Run a rights audit on the records.
  2. Check customer consent and confidentiality clauses.
  3. Apply de-identification before licensing.
  4. Remove data that cannot be licensed.
  5. Negotiate the license agreement clauses.

Buyer and Law Pages Named on This Page

Each program named above has its own page with dates and sources.

Frequently Asked Questions

Can a bank license customer data to AI companies?

Only within the rules that bind each record: for consumer customers, notice and a chance to opt out come before disclosure to a nonaffiliated third party (15 U.S.C. 6802(b)(1)), and commercial customer files turn on contract and other law because Regulation P does not cover them (12 CFR 1016.1(b)(1)).

Does Swiss bank secrecy stop a Swiss bank from licensing data?

Article 47 of the Banking Act punishes disclosure of an entrusted secret, and its text does not mention anonymized data, so a Swiss bank checks it before any license.

Does DORA apply to a data license to an AI developer?

Its text covers a financial entity's use of ICT services and does not address licensing data out. It reaches the tools and providers around a license through the contract terms of Art. 30(2).

How much do AI companies pay banks for data?

No bank-specific price is on record in the sources checked for this page. Mercor states USD 80,000 to 900,000 for its Payments & Fintech listing as of 2026-11-17, a claim for payment and commerce firms and not for banks. Valuation methods are set out under what company data is worth.