Ethics, governance and professional practice

PUBLISHED

Obligations to other people arising from using AI at work.

Statements

—

Version

v1.29

Last reviewed

30.08.2026

In scope

Identifying who may be affected, and how

Bias and fairness as operational practice rather than principle

Transparency and disclosure: when the use of AI must be stated

Attribution, authorship and intellectual property

Consent where AI is applied to another person's information or work

Accountability, and who answers for a decision AI influenced

Regulatory landscape, and recognising where an obligation attaches

Organisational policy: following it, and contributing to it

Escalation, and refusal to proceed

Documentation as a professional duty

Environmental and resource considerations

Out of scope

Excluded	Where it lives
Legal advice	`Out of framework`
Jurisdiction-specific compliance detail	`Indicators note that specifics are local`
Assurance evidence itself	`D6`

Boundary notes

Where this domain abuts another, and how the line is drawn

Against D5 Solution Design

D5 covers whether AI is suitable. D7 covers whether its use is permissible.

Against D6 Evaluation and Assurance

D6 produces evidence. D7 covers the obligation to produce it and to act on it.

Against regulation

D7 statements must survive changes in regulation. Anything that would be invalidated by a single jurisdiction amending a statute is written at the wrong level of abstraction.

Statements by level

Every identifier is a permanent address. Indicators are normative; they state what would be observed in a person who meets the statement.

L1 Aware

4 statements

D7.L1.01Knowledge

State that using an AI system does not transfer responsibility for the result.

Indicators

States who remains accountable for AI-assisted work

Recognises that a tool’s error is still the user’s output

Rejects the system as an explanation for a wrong result

D7.L1.02Judgement

Identify where AI use must be disclosed.

Indicators

Names the situations in their own work requiring disclosure

States that concealment is the harm, not the use

Asks rather than assuming disclosure is unnecessary

D7.L1.03Knowledge

Recognise that AI output can affect people unequally.

Indicators

States that a system can work well on average and badly for a group

Identifies a decision in their work where that would matter

Reports rather than dismissing an apparent pattern

D7.L1.04Knowledge

State the obligations that apply to their own AI use.

Indicators

Names the policy, professional and legal obligations that apply

Identifies where the obligations are stricter than the tool’s defaults

Knows where to ask when an obligation is unclear

L2 Applied

6 statements

D7.L2.01Practice

Disclose AI involvement in work where disclosure is required.

Indicators

Discloses at the point the work is delivered, not on request

States what the system did rather than that it was used

Applies the organisation’s form of disclosure

D7.L2.02Practice

Apply organisational and legal obligations to a specific AI task.

Indicators

Identifies the obligations engaged by the task before starting

Applies the stricter obligation where two conflict

Refers rather than resolving a conflict alone

D7.L2.03Skill

Check AI-assisted work for effects that fall unevenly on a group of people.

Indicators

Checks output across the groups the decision affects

Records what was checked and what was found

Reports a pattern rather than correcting individual cases

D7.L2.04Practice

Attribute and respect the rights in material used and produced.

Indicators

Establishes rights before publishing or reusing

Attributes as the licence requires

Refers material whose rights are unclear

D7.L2.05Practice

Raise a concern about an AI use they believe to be wrong.

Indicators

Uses the route the organisation provides

States the concern in terms of consequence and evidence

Escalates where the first response does not address it

D7.L2.06Skill

Explain to an affected person, accurately, how AI was involved in a decision.

Indicators

Explains in terms the person can act on

Does not overstate or understate the system’s role

States the route to challenge the decision

L3 Proficient

8 statements

D7.L3.01Judgement

Assess the impact of a proposed AI use on the people it affects.

Indicators

Identifies affected parties, including those who are not users

States the harms considered and the evidence for the conclusion

Records the assessment before the decision is taken

D7.L3.02Skill

Design fairness requirements for a system and the evidence they require.

Indicators

States the fairness property being sought and its limits

Defines measurement that could show the requirement is not met

Provides for the trade-off between competing fairness properties

D7.L3.03Skill

Establish transparency and explanation requirements for a system.

Indicators

States what must be explained, to whom, and to what depth

Distinguishes an explanation from a description of the mechanism

Provides for the case where no adequate explanation exists

D7.L3.04Judgement

Advise on a use that is permitted but should not proceed.

Indicators

Separates the legal question from the question of whether to act

States the ground in terms of consequence to identifiable people

Records the advice, including where it was not taken

D7.L3.05Skill

Design the route by which an affected person can challenge an AI-influenced decision.

Indicators

Provides a route that reaches someone with authority to change the outcome

States the timescale and what the person must supply

Requires the outcome of challenges to be recorded and reviewed

D7.L3.06Judgement

Assess the labour and skill effects of an AI deployment.

Indicators

Identifies work displaced, degraded or intensified

Consults those affected rather than reporting on them

States what would mitigate the effect and what it would cost

D7.L3.07Judgement

Apply the obligations of a regulated or professional context to AI use.

Indicators

Identifies the obligations that survive the introduction of a system

States where the tool cannot be used within them

Documents the basis for continued use

D7.L3.08Judgement

Review an AI-related incident for the ethical failure, not only the technical one.

Indicators

Identifies the decision, not only the defect, that produced the harm

States who was accountable and what they were expected to do

Produces a change in practice, not only in the system

L4 Advanced

4 statements

D7.L4.01Practice

Establish the organisation’s principles for AI use, in terms that decide cases.

Indicators

States principles that can rule a proposal out

Provides the route by which a hard case is decided

Publishes the principles to those affected, not only internally

D7.L4.02Practice

Define accountability for AI-influenced decisions across the organisation.

Indicators

Assigns named accountability rather than collective ownership

States what accountability requires the holder to do

Provides for the accountable person to say no

D7.L4.03Practice

Establish how concerns and dissent about AI use are heard and acted on.

Indicators

Provides a route that does not depend on the goodwill of the subject of the concern

Records concerns and their outcomes

Protects those who raise them

D7.L4.04Judgement

Account publicly for the organisation’s AI use, including where it went wrong.

Indicators

States what is disclosed, to whom, and on what cycle

Discloses failures with the same specificity as successes

Does not treat commercial discomfort as a reason to withhold

Editorial notes

Known gaps, open questions and contested points

Published because the record is more useful than the appearance of completeness.

Drafted last, deliberately. D7 depends on vocabulary established across all seven other domains, and drafting it early produces general statements that are rewritten anyway.

It is also the domain most vulnerable to becoming a list of principles rather than a set of observable competences. Every statement must survive the test: what would I watch this person do?

All eight

Domains index

Filtered view

Statements in D7

Proficiency

Levels

This page displays version 1.29 · last reviewed 30.08.2026