Resource Library · November 2026

Enterprise AI Governance Framework

A design framework: fourteen tests, a proportionate path, and a first case that raises the governance. Not a compliance certificate, not legal advice, and not a finding that a system is aligned with any named regime.

22 pages35 min readAdvancedCompanion workbook included

By AsymiLink Meta · Published · Updated

Request the guide and workbook

Summary

A design framework: fourteen tests, a proportionate path, and a first case that raises the governance. Not a compliance certificate, not legal advice, and not a finding that a system is aligned with any named regime. Frameworks excerpted on this page: Fourteen design tests (excerpt); A proportionate path, not a badge. A 22 pages November 2026 document, with a companion workbook.

  • Fourteen design tests, applied to every case, including a document assistant and a system that affects a claim.
  • Consequence determines the depth of the path. This framework will not make a legal classification.
  • Counsel is asked in time, with written questions. This document is not that advice, and it will not recite a date.
  • Fund a first case that raises the governance. A first case that skips the inventory “to show value” has skipped the design.

A crawlable plain-text copy of this page is also at /resources/ai-governance-framework.txt. The canonical URL is https://asymilinkmeta.com/resources/ai-governance-framework.

Written for Risk, legal, technology and executive sponsors who have to put a governance path under a first AI case, and under the systems that are already running.

What's in this framework

This 22-page document is a design framework for putting a governance path under a first AI case, and under the systems that are already running. The library has sometimes described it as a comprehensive model for regulatory alignment, including named regimes, a bias-testing protocol with thresholds, and a prescribed committee. It is not a compliance checklist that closes the law. It does not implement or certify alignment with any named regime. It does not publish a bias-testing threshold. It does not prescribe a committee as if one structure were required.

What it is, instead, is fourteen tests to apply before a first case is funded; a way to treat inventory, proportionate governance, effective challenge, an operator, a switch-off and an audit trail as architecture; and questions for counsel — including, in qualitative terms, the risk-based structure of regimes your counsel may already be considering. What applies, and when, is a question for your counsel. This document is not legal advice, and it will not recite a date.

It is written for risk, legal, technology and executive sponsors. Two illustrative institutions — the same mid-size insurer and the same community bank as in the 2026 diagnostic — are used to show how the tests would read, today. They are labeled illustrative. They are not client case studies. A companion workbook holds the tests, a sheet for obligations to name, and the two institutions. A completed workbook is not a certificate.

Proportion, counsel, and a first case that raises the path

Consequence determines depth. A document assistant and a system that would affect a claim or a credit decision are held to the same tests; the evidence and the gates are deeper where the consequence is higher. This framework will not make a legal classification, and it will not assign a risk tier as if it were a regulator.

Counsel is asked in time. Eight questions, not a complete list, and not advice, sit in Section 3. A missing answer is a failed test, not permission. The document will not be used as a calendar, and it will not close a question that belongs to counsel.

A first case should raise the governance, not skip it. Start with what is already running, put it on the inventory, and fund work that makes the operator, the switch-off, the sampling and the audit trail real. Do not wait to be “Leading” on every dimension. The readiness assessment in this library is the companion for that position. This framework will not stand in for it, and it will not issue a certificate that a system is aligned with any named regime.

Fourteen design tests (excerpt)

The same standard for a document assistant and for a system that affects a claim. A failed test on inventory, counsel, the gate, or the first case is usually a reason to pause.

  1. T1

    Every system that is running, or is about to run, is on an inventory the organization already uses — or a written decision says why not, and who stands behind that.

  2. T7

    A person can stand behind a decision that affects someone. The gate is architecture. This framework will not describe an unsupervised high-consequence system as a first case.

  3. T8

    Counsel has been asked, in time, what applies. A missing answer is a failed test, not permission. This document is not that advice, and it will not recite a date.

  4. T10

    The first case is bounded, and raises the governance. A first case that skips the inventory “to show value” has skipped the design.

  5. T11

    People who are affected can be told, honestly, what the system does. What to tell them, and when, is a design and a counsel question.

  6. T12

    Fairness is tested for this use, not scored against an invented threshold. A frozen set and a person who can stand behind the examination.

The complete fourteen tests — including operator, switch-off, audit trail, training and reuse, and proportionate depth — are in the PDF and the workbook.

A proportionate path, not a badge

Section 4 treats proportion and risk as a design. Consequence determines depth. The framework will not make a legal classification.

  1. Inventory first

    What is already running is on the inventory, or a written decision says why not. Assistants people already use are not “informal” and therefore off the path.

  2. Depth follows consequence

    The tests are the same. The evidence, the gates and the sampling are deeper where a decision affects a person. That is a design, not a badge.

  3. A first case that raises the path

    Fund work that makes the operator, the switch-off and the audit trail real. Do not wait to be Leading. A convenient path that skips T10 is a failed test.

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.

By requesting this resource you agree that AsymiLink Meta may contact you about it. See our Privacy Policy and Terms of Use.

Frequently asked questions

Does this framework certify alignment with a named regime?

No. It is a design framework, not a compliance certificate and not a finding that a system is aligned with any named regime. A completed workbook is not a certificate.

Is this legal advice?

No. Counsel is asked in time, with written questions. This document will not close those questions, and it will not recite a date by which a rule applies.

How does this relate to the AI readiness assessment?

The readiness assessment is how you take a position on eight dimensions, including governance. This framework is the design path under a case. A first case should raise the governance, not wait for a top score on every dimension.

What is in the companion workbook?

The fourteen tests, a sheet for obligations to name, and two illustrative institutions. A completed workbook means the tests were asked. It is not a certificate.