Data and information

PUBLISHED

Sourcing, judging and governing the information AI systems consume and produce.

Statements

—

Version

v1.29

Last reviewed

30.08.2026

In scope

Fitness for purpose of a data source for a given task

Provenance, lineage and the ability to say where information came from

Personal and sensitive data in prompts, context and logs

Consent and lawful basis, described abstractly with jurisdictional specifics in the indicators

Retention, minimisation and deletion in AI workflows

Structured and unstructured sources, and their different failure modes

Grounding output in sources, and citation of them

Verification of AI-produced claims before reliance

Information hygiene in shared and persistent AI workspaces

Out of scope

Excluded	Where it lives
Data engineering pipelines and warehouse architecture	`Out of framework`
Database administration	`Out of framework`
Retrieval and grounding architecture in depth	`X-RAG`

Boundary notes

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

Against D8 Security

D2 covers judgement about data. D8 covers protecting it. Where a statement concerns an adversary, it is D8.

Against D6 Evaluation and Assurance

D2 covers whether a source is fit to use. D6 covers whether the resulting system is fit to rely on.

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

D2.L1.01Knowledge

State where the information an AI system uses comes from.

Indicators

Distinguishes what the system was trained on from what it was given

Identifies whether the system can retrieve anything at the time of use

Recognises that the person supplying material is part of the supply

D2.L1.02Judgement

Recognise that information entered into a system may travel beyond the immediate task.

Indicators

States that prompts and attachments may be retained or logged

Checks before entering material whose sensitivity is unclear

Applies the same caution to context and history as to the prompt

D2.L1.03Judgement

Identify information that requires a source before it is used.

Indicators

Distinguishes a figure, date or quotation from a paraphrase

Identifies claims that would cause harm if wrong

Seeks the source rather than the system’s restatement of it

D2.L1.04Knowledge

Describe why an AI system may present unsourced content as though it were sourced.

Indicators

States that citations may be generated rather than retrieved

Checks that a cited source exists and says what is claimed

Recognises a plausible reference as no evidence at all

L2 Applied

7 statements

D2.L2.01Skill

Check a factual claim in AI output against an authoritative source.

Indicators

Identifies the authoritative source for the class of claim

Checks the claim, not merely the existence of the citation

Records what was checked and what was not

D2.L2.02Skill

Cite the sources used in and produced by AI-assisted work.

Indicators

Cites the underlying source rather than the system

States where the system contributed and how

Applies the organisation’s citation convention consistently

D2.L2.03Practice

Apply classification rules to material before using it with an AI system.

Indicators

Classifies before entering rather than afterwards

Applies the rule to attachments, context and retrieved material

Refers material whose classification is unclear

D2.L2.04Skill

Establish the licence and permitted use of material supplied to or produced by a system.

Indicators

Identifies the licence before use rather than after publication

States whether the intended use is permitted

Refers cases where the licence is silent or unclear

D2.L2.05Skill

Prepare source material so that a system can use it reliably.

Indicators

Removes duplication, obsolete versions and contradictions

Preserves structure the task depends on

States what was excluded and why

D2.L2.06Skill

Record the provenance of AI-assisted output, including the material supplied.

Indicators

Records which sources were supplied and which were retrieved

Records the date and version of each source

Retains the record where the output is used

D2.L2.07Skill

Recognise output shaped by an unrepresentative or incomplete source set.

Indicators

Identifies material absent from the supply that the task required

Distinguishes an omission in the source from an error in the output

Reports the gap rather than compensating for it silently

L3 Proficient

8 statements

D2.L3.01Skill

Design the information supply for an AI system, including sources, refresh and exclusions.

Indicators

States each source, its owner and its refresh interval

States what is deliberately excluded and why

Provides for a source becoming stale or unavailable

D2.L3.02Judgement

Assess whether a source set is fit for the questions it will be asked.

Indicators

Tests the set against the questions actually expected

Identifies question classes the set cannot answer

States the consequence rather than only the gap

D2.L3.03Skill

Diagnose failures caused by source quality rather than by the model.

Indicators

Isolates the source contribution before changing the model or instruction

Traces an incorrect output to the material that produced it

Corrects the source rather than patching the output

D2.L3.04Skill

Design retention, minimisation and deletion for AI workflows.

Indicators

States what is retained, where, and for how long

Minimises what is supplied to what the task requires

Provides a deletion path that reaches context, logs and derived copies

D2.L3.05Judgement

Assess bias arising from source composition and state its consequence.

Indicators

Identifies whose material is over- and under-represented

States the effect on specific decisions rather than in general terms

Distinguishes a source bias from a model bias

D2.L3.06Skill

Establish provenance requirements for output that will be published or relied upon.

Indicators

States the evidence required before publication

Defines who verifies it and how the verification is recorded

Provides for output whose provenance cannot be established

D2.L3.07Judgement

Advise on the use of personal, confidential or licensed material with AI systems.

Indicators

States the basis on which the use is or is not permitted

Identifies where an alternative achieves the purpose with less exposure

Records the advice and its reasoning

D2.L3.08Skill

Design how conflicting sources are resolved and how the resolution is recorded.

Indicators

States precedence between sources in advance

Surfaces unresolved conflict rather than selecting silently

Records which source was preferred and why

L4 Advanced

4 statements

D2.L4.01Practice

Establish the organisation’s standard for information used by AI systems.

Indicators

Defines minimum provenance, quality and licensing requirements

Sets requirements proportionate to consequence

Establishes how compliance is verified

D2.L4.02Practice

Define ownership and accountability for source sets and their maintenance.

Indicators

Assigns a named owner to each significant source set

States the maintenance obligation, not only the ownership

Provides for the owner’s departure

D2.L4.03Judgement

Set the organisation’s position on training, retention and reuse by third-party providers.

Indicators

States what may and may not be shared with a provider

Establishes how contractual terms are verified rather than assumed

Reviews the position when terms change

D2.L4.04Practice

Hold accountable those responsible for source quality where output has caused harm.

Indicators

Traces the harm to the source and the obligation that was not met

Treats absent provenance as a finding rather than a neutral state

Requires a change to the supply, not only to the output

Editorial notes

Known gaps, open questions and contested points

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

Verification of AI-produced claims is the single highest-value competence in this domain and is under-taught everywhere. It should carry statements at L2, L3 and L4.

All eight

Domains index

Filtered view

Statements in D2

Proficiency

Levels

This page displays version 1.29 · last reviewed 30.08.2026