Written for Privacy, security and risk owners at community banks and credit unions, and the sponsors who have to ask design questions about AI that touches customer information.
What's in this checklist
This checklist is the operational document for design questions about AI that touches, or might touch, customer information, under the heading of the GLBA Safeguards. It is complementary to the Financial Services AI Playbook, the community-banks first-case playbook, the LLM Security white paper, and the SOC 2 readiness checklist already in this library, and is not a replacement for them. This checklist will not reprint their tests, will not provide payloads, and will not treat a completed SOC 2 sheet as a close of GLBA. It is design questions only. It is not legal advice. It will not close GLBA. It will not claim that a system "meets GLBA." It will not recite dates. That is the same stance this library takes on a named health-privacy framework: we will not certify that a system meets a named regime.
What it is, instead, is a design-question checklist: six families of questions, fourteen design tests, and a first-case path consistent with the finance industry playbook. The first work is an inventory of AI that touches, or might touch, customer information — whether it does is a counsel question this checklist will not close — and then design questions about what information, for what purpose, about access and safeguarding evidence, about vendors, about a trail, and about hosting. A completed sheet is not a finding that a system meets GLBA, and not a close of any named regime. A vendor page is not a GLBA close. Isolation is not a GLBA close. Retention is a counsel question; this checklist will not recite a period. A twelve-week "GLBA program" is not a first case this checklist will recommend. This checklist will not invent a pass rate, and will not get the reader a finding that a system meets GLBA.
How the evidence tests and the method are meant to be used
The financial method is The Complete Guide to AI Automation ROI, where a process is being funded. This checklist will not invent a GLBA-sector return, and it will not promise a twelve-week program. The 3.8x average first-year ROI across client engagements is an average, not a forecast or a commitment for any individual project. ROI is measured across client engagements against a baseline agreed before work starts. It is an average, not a forecast for any individual project, and is not used as a sector benchmark. Built to SOC 2 standards is the exact wording we use; it is not a certification of the reader's system, and not a GLBA close.
The illustrative firm is the same hypothetical community bank as in the Financial Services AI Playbook, the community-banks first-case playbook, the 2026 diagnostic, and the financial-services practitioner's framework: overall 22, Developing, with a next-step target of 25, still Developing, Governance raised from 3 to 4 so that it covers what is actually running. Two generative-AI assistants have not been added to the model inventory. Whether those assistants touch customer information is a counsel question this checklist will not close. The honest first work is still to put those assistants on the inventory, with an owner, a sampling plan and a stop, and only then decide whether a fraud-triage assist is a different project — the same first case as in the finance industry playbook. The firm is illustrative, not a sample. The target is a next step, not a timeline, not a promise, and not a finding that a system meets GLBA. A companion workbook, asymilink-glba-safeguards-checklist.xlsx, holds a family register, the fourteen tests, questions for counsel, a baseline sheet, and the same illustrative bank. Do not wait to be Leading. Do not treat a completed sheet as a close of GLBA.
Six families of design questions
each with an honest first case, a case that is too wide, and a gate. Each item below is the shape of the map; the full wording, honest first cases and “too wide” warnings are in the PDF.
F1
Inventory of AI that touches, or might touch, customer information (first; whether it does is a counsel question this checklist will not close)
F2
What information, for what purpose
F3
Access and safeguarding evidence, not a “meets GLBA” finding
F4
Vendor and subprocessor evidence (a vendor page is not a GLBA close; we will not claim a certified connector)
F5
A trail that can reconstruct a day, with retention a counsel question and no period recited
F6
Hosting, asked as a design question (isolation is not a GLBA close)
The PDF gives, for each family, an honest first case, a case that is too wide, and a defined gate. The summaries above are the shape of the map, not a substitute for that wording.
Fourteen evidence tests (excerpt)
Fourteen evidence tests are applied to every case. The answers will differ. The standard does not. The full wording is in the PDF and the companion workbook.
T1–early
a named inventory; counsel asked without a date taken from this document; a refusal to claim that a system meets GLBA.
Mid-tests
a refusal to treat a completed sheet as a close, or to recommend a twelve-week “GLBA program” as a first case; vendor and hosting asked as design questions.
Later tests
a signed baseline on the ROI method where a process is being funded; isolation is not a GLBA close.
A completed sheet is not a finding that a system meets GLBA, and not a close of any named regime.
Get the guide and workbook
Enter your name, work email and company. The files appear on this page immediately. A backup download link is also emailed when email delivery is available.