Can we tell the person affected why, in a way they can actually understand?
Enhanced: 22 controls
Who it applies to
Enter for every AI system whose letter is E or A (this area has no Baseline controls). Go as deep as your system letter, and implement in your mode.
Why it exists
When an AI decision affects a person, be able to explain it to that person — in their terms, not the model’s. An explanation that only a data scientist could follow is not an explanation to the customer who was refused, the applicant who was rejected, or the patient whose care was shaped; and in a growing number of jurisdictions, people have a right to a meaningful account of a decision made about them by automated means, and a right to contest it. Build the capacity to give that account before you deploy, because you cannot retrofit it onto a decision already made.
What to actually do
Produce an explanation the affected person can act on. Tell them the main factors that drove the outcome, in plain language: not the model’s internal weights, but the reasons a person could understand, check against their own knowledge, and challenge if they are wrong. The test is whether the person could use the explanation to contest the decision — if they could not, it is not yet an explanation.
Preserve the right to challenge, and route it to a human. An explanation without a way to contest it is a courtesy, not a safeguard. Give the affected person a route to challenge the decision and have it reviewed by a competent human — which connects to the redress mechanism in the Accountability domain and the human oversight of its own domain.
Controls
Enhanced2 controls
Control 167
Be able to explain any person-affecting AI decision to the person, in plain terms.
EnhancedPreventiveall rolesper system
What it is
The main driving factors, understandable and checkable by the affected individual.
How to implement it · Minimum structure
The decision, and that AI was involved in reaching it
The two to four main factors that drove the outcome, in plain language
The data about the person that was used, so they can check it
What would have changed the outcome, where this can be stated
How to contest it and reach a human reviewer
Before use: tested with a few non-expert readers who can restate why
How to start
Draft an explanation template for the most frequent person-affecting decision and test it with three people outside the team.
Evidence
The explanation templates, samples of explanations issued, and the comprehension test notes.
Mode: light / standard / full
light
The reviewer writes each explanation from the template, using the written criteria and the factors noted during review.
standard
The system outputs its main factors per case and staff turn them into the plain-language template.
full
Explanations are generated per decision from recorded factors, with sample quality checks by a dedicated team.
Control 168
Provide a route to contest a decision and have a human review it.
EnhancedPreventiveall rolesper system
What it is
Connected to redress and human oversight.
How to implement it · Steps
Channel: a named contact point (email, form or phone) stated in every decision notice
Intake: log each challenge with date, decision concerned and reasons given
Review: by a competent person who did not make the original decision and can change it
Response time: set and communicated (e.g. within 30 days)
Outcome recorded and fed into the redress mechanism and the oversight review of the system
How to start
Name the reviewer and the contact point, add them to the decision notices, and open a log for challenges.
Evidence
The challenge log with outcomes and the identity of the human reviewer for each case.
Mode: light / standard / full
light
A named manager handles challenges from a shared mailbox and logs them in a spreadsheet.
standard
A designated review team handles challenges within the set deadline and reports outcomes to the system owner each quarter.
full
A case-management workflow routes challenges to an independent review function and feeds trends into system oversight.
Evidence
Sample explanations tested for lay comprehension; the contestation route and records of its use; evidence of human review on challenge.