Start here · Free Director’s Guide · Six pages
The Director's Guide to The Assurance Line
Understand the eight domains, how evidence supports assurance, and which resources to use for your board's work. Read below or download a copy to keep. No sign-in required.
Start here · A guide to the framework
The Director's Guide to The Assurance Line
Understand the eight domains, how evidence supports assurance, and which resources to use for your board's work.
What the framework helps you establish
An AI update may tell you what the organization has launched, what it hopes to achieve, and which policies it has approved. As a director, you also need to know what supports those statements and what remains uncertain.
The Assurance Line organizes that conversation around eight questions, asked in order. They take you from knowing what AI is in use to knowing how the organization would respond when it goes wrong.
The framework connects those questions to requirements, supporting evidence, and five assurance states. Its companion resources help you prepare for a meeting, examine management’s answers, and follow up on gaps.
You do not need to become an AI engineer to use it. You need to distinguish what management can demonstrate from what it has described, promised, or assumed.
This guide explains the structure and where to start. Use it alongside the framework overview; turn to the detailed Board Guide when you need help with decisions, follow-up questions, and monitoring in each domain.
The useful outcome is a clearer account of what the organization knows, what supports that knowledge, and what needs attention next.
Understanding the framework · 02
Why the eight questions follow a sequence
Each domain answers a question that the later domains depend on. Start with Visibility: what AI is actually in use? Then establish who answers for it, why it is used, and what it could do to the organization.
The later questions examine what the systems are built on, which obligations apply, whether the claims have been independently tested, and how failures would be detected and handled.
The sequence matters because a convincing answer can rest on an uncertain foundation. Independent testing of the systems on a list cannot establish whether that list includes every system in use.
You can investigate a particular issue at any point. When you report the result, make the underlying scope and dependencies clear. A strong result later in the line does not repair a gap earlier in it.
- 01VisibilityWhat AI are we actually running?
- 02AccountabilityWho is answerable for it?
- 03ValueWhy are we running it?
- 04ExposureWhat could go wrong, and who could be affected?
- 05FoundationsWhat is it built on?
- 06ObligationsWhat are we required to do?
- 07AssuranceHow do we know any of this is true?
- 08ResilienceWhat happens when it goes wrong?
Understanding the framework · 03
How to read the framework
Every domain uses the same structure. Purpose and scope explain what it addresses and which systems or activities it covers. Reasoning explains why the requirements matter.
Control objectives state what needs to be in place. Each has an identifier, such as VIS-1, that you can follow across the framework and its companion resources. Supporting evidence names records or artifacts to examine and explains what their contents, or absence, may reveal.
Assurance criteria describe what each of the five states looks like in that domain. They help distinguish a documented arrangement from one that has been demonstrated and independently tested.
The standards mapping connects requirements to selected external references. It provides context and traceability. It is not certification, a finding of legal compliance, or a complete account of every obligation that may apply. Neither a downloaded guide nor a completed self-assessment supplies those conclusions.
An objective is a requirement, not a box to tick because a document exists. The example below comes from Visibility, which is available in full for free.
Reading a requirement · Visibility
- The control objective
- TAL 1.1 / VIS-1
A single inventory of AI systems in production is maintained. - The supporting evidence
- The inventory, with its last-verified date and the name of the person who verified it.
- The question for your discussion
- What supports the inventory’s accuracy, and how would management find a system missing from it?
Understanding the framework · 04
What an assessment result means
The five assurance states describe how much confidence the evidence supports. They are shared across all eight domains; the detailed criteria explain how each state applies to a particular domain.
A self-assessment records the organization’s own answers. It can reveal gaps and direct follow-up work, but completing it does not establish independent assurance. Selecting Assured does not perform or replace the testing that state requires.
Read the domain results together with their scope, supporting evidence, and unanswered items. An incomplete assessment remains incomplete; it does not establish a result for the whole organization.
- 1Absent
Nobody is doing this.
- 2Ad hoc
It happens sometimes, and depends on particular individuals.
- 3Defined
Documented, assigned, and repeatable.
- 4Evidenced
Produces artifacts a third party could test.
- 5Assured
Independently tested, reported to the board, and acted on.
Results are never averaged.
Each domain stands on its own result. A strong score in one does not make up for a weak one in another, and no overall score is given.
Nothing is more assured than what it rests on.
A domain’s result is capped at the lowest result of any domain before it on the Line. If the inventory is only Ad hoc, a claim that most systems are independently tested rests on a count nobody can trust, and the result says so.
Understanding the framework · 05
Where you and management fit
The framework supports the board’s oversight and management’s work. Use your organization’s delegation arrangements to establish who decides what and which matters return to the board.
Your role as a director
Ask whether material AI uses are understood, responsibility is clear, and management’s claims are supported. Examine the limits of the evidence and whether important gaps have an owner and a follow-up date.
You can challenge an assurance claim without inspecting every technical test. Ask what was tested, by whom, what was outside the scope, and what happened after the findings were reported.
Management’s role
Identify the systems in use, assign responsibility, establish controls, and provide evidence that they operate. Bring material decisions, unresolved exposure, and incidents to the board under the agreed arrangements.
Management also needs to keep the picture current as systems, data, vendors, and uses change. An inventory or assessment from an earlier meeting may no longer describe the organization’s AI.
Close the conversation with a next step
For an unresolved issue, record the decision or evidence needed, the person responsible, and when the board will hear back. A useful discussion should leave management with a clear action and the board with a clear expectation.
Understanding the framework · 06
Choose the resource for your next task
Start with the work you need to do. The free resources explain the approach and give you a starting point. The paid editions provide the detailed resources for applying it.
The Board Edition focuses on meeting preparation and examining management’s answers. The Professional Edition includes the board resources and adds the complete framework, full assessment, and practical governance tools.
Get up to speed on AI oversight
Free
Explore the approach and a worked example
Free
Start a discussion about gaps
Free · Eight questions
Prepare for a board discussion
Board Edition · Also in Professional Edition
Review a project or release decision
Professional Edition
Apply requirements, assess gaps, and report
Professional Edition