SETTING GLOBAL STANDARDS FOR TRUSTED AI CREDENTIALSAI Competence Framework v1.29 · current release
You are reading the current version of the framework, v1.29.Permanent address for this version

For employers

Write requirements against domains and levels instead of tool names, and you can still recruit for them in two years. Everything on this page is copyable and needs no permission.

Requirements that survive

Tool names date. Competences do not.

✗ 3+ years using [named AI product] in production
✓ D4 Automation and integration at L2, evidenced in any toolchain
✗ Prompt engineering expert
✓ D3 Interaction and elicitation at L3: adapts approach when a first attempt fails, and explains why
✗ Strong understanding of AI risk
✓ D6 at L3 and D8 at L2: designs evaluations, reports their limits, identifies adversarial exposure
✗ AI certification required
✓ Competence at the levels listed; a credential mapped to the framework is one way to evidence it, not the only one
Choosing the level

Which level the role actually needs

LevelThe role needs it whenTypical
L1The person uses AI output but does not decide anything on it alone, and escalates rather than resolves.Most roles
L2The person produces work with AI in familiar situations, to an established procedure, and documents it.The working majority
L3The person faces non-routine situations with competing constraints, chooses an approach, and answers for the choice.One or two domains per senior role
L4The person sets the standard others follow and is accountable for outcomes at organisational scale.Rare; heads of function
Job description fragments · copy and paste

One fragment per level, ready to drop in

L1

Any role using AI tools incidentally

You will work alongside AI systems in this role. We expect competence at level L1 (Aware) in domain D6 (Evaluation and assurance) of the AI Competence Framework v1.29: you can describe why AI output requires evaluation and identify when to escalate. No credential is required; we assess this in conversation at interview.

L2

Practitioner using AI in daily work

Required: AI Competence Framework v1.29 at level L2 (Applied) in D6 (Evaluation and assurance) and D2 (Data and information). You will produce evaluation records for outputs you rely on. A credential is welcome but not required.

L3

Senior practitioner or lead

Required: AI Competence Framework v1.29 at L3 (Proficient) in D6 (Evaluation and assurance), with L2 in D8 (Security). At L3 you will design evaluation approaches for non-routine work, justify the trade-offs, and review the work of others.

L4

Head of function, standard-setter

Required: AI Competence Framework v1.29 at L4 (Advanced) in D7 (Governance and accountability) and L3 in D6. At L4 you will set the standard others work to and hold accountability for outcomes across the function.

Internal capability matrix

What your teams hold, and where the risk is

RoleD1D2D3D4D5D6D7D8
AnalystL2L2L2L1L2L1L1
EngineerL2L3L2L3L3L2L1L3
Product managerL2L2L2L1L2L2L2L1
Team leadL2L2L2L2L2L3L3L2
Head of functionL2L3L2L2L3L3L4L3
Assurance reviewerL2L3L2L2L2L4L3L3

Illustrative targets, not a recommendation. A dash means the role does not need the domain.

Interview and evidence prompts

Questions derived from the indicators, not from the statement text

D6.L2.05

Record the basis on which an AI output was accepted for use.

Ask

Show me an evaluation record you wrote. Talk me through it.

What did you write down that you were tempted to leave out?

Listen for

A real artefact, and an account of a decision that was uncomfortable.

D6.L3.04

Report the limits of an evaluation to the person relying on it.

Ask

Tell me about a time you had to tell someone your evaluation could not answer their question.

How did they respond, and what did you do next?

Listen for

Evidence they said it out loud to someone with more authority than them.

D8.L3.01

Diagnose how untrusted content reached a system's instruction path.

Ask

Walk me through an injection path you found, in any system.

What did you change, and how did you confirm it worked?

Listen for

A specific path, a specific fix, and a verification step.

D2.L2.04

Apply the organisation's data-handling rules to material entering an AI system.

Ask

What have you refused to put into an AI system, and why?

Who did you check with, and what did they say?

Listen for

A concrete refusal.

A caution, offered because it will happen anyway

Using the framework for performance management is a different instrument

The framework describes competence: what a person can do. Performance management is about what a person did, with particular support and particular constraints.

USETo identify what development a person needs, and to agree what they will be able to do afterwards.
USETo define a role before recruiting, and to structure interview evidence consistently across candidates.
USETo plan capability across teams, where the unit of analysis is the team rather than the person.
AVOIDAs a scoring rubric in appraisal, pay or promotion decisions. Levels describe how a person works, not how well they performed.
AVOIDAs a basis for capability procedures or dismissal. The framework has no assessment methodology for individual performance.
AVOIDAs a self-assessment that feeds an HR system unchanged.