The OECD AI Principles, and why they keep reappearing.
They bind nobody directly and they are behind a great deal of what does bind you. National AI strategies, regulator guidance and procurement questionnaires across dozens of countries were drafted with these in the room — which makes them the cheapest common denominator a multi-jurisdiction programme can organise around.
What are the OECD AI Principles?
An intergovernmental standard on artificial intelligence, adopted by OECD member countries and adhered to by others. They are values-based rather than prescriptive — they describe what trustworthy AI looks like and leave the mechanism to each jurisdiction, which is precisely why they turn up underneath so many national frameworks.
- Status Non-binding A recommendation adhered to by governments, not a statute binding on you.
- Shape Values-based Principles for AI actors, plus recommendations for national policy.
- Reach Wide by reference Referenced in national strategies and regulator guidance across many jurisdictions.
- Best used as A foundation The layer other frameworks sit on, not a substitute for any of them.
Who uses them?
Mostly organisations that need one story to tell in several places at once.
- Multinational programmes One set of principles that maps onto national frameworks in most of the places you operate, rather than a separate narrative per country.
- Organisations answering questionnaires Procurement and diligence questions are frequently drafted from this vocabulary even when the framework is never named.
- Teams starting from nothing A defensible place to begin when the binding framework is still unclear, because almost every candidate framework is compatible with it.
- Public-sector suppliers Government AI policy in many countries is written on this foundation, and the terms inherit the vocabulary.
How iDharma supports OECD alignment
Alignment is only worth claiming if it can be evidenced. That is the work.
| Principle area | What the assessment does |
|---|---|
| Inclusive growth and wellbeing | Whether benefit and harm were considered for the people affected, not only for the operator, and whether that consideration is on the record. |
| Human-centred values and fairness | Testing for discriminatory outcomes, and whether human rights considerations reached the design rather than the launch review. |
| Transparency and explainability | What is disclosed, to whom, and whether an explanation would mean anything to the person receiving it. |
| Robustness, security and safety | Testing, monitoring and the ability to intervene or withdraw a system that starts behaving differently. |
| Accountability | Named owners, real authority, and a trail from a finding to a change. |
| Cross-framework mapping | Each finding tagged to its NIST AI RMF category, ISO 42001 clause and EU AI Act article, so alignment is evidence rather than assertion. |
The Principles are a foundation, not a destination. If a binding framework reaches you — NIST AI RMF, ISO 42001, the EU AI Act — build to that and use this to explain how the pieces relate.
Five principles for AI actors
Values-based, which makes them portable and makes evidencing them the hard part.
Inclusive growth and wellbeing
Who benefits, and who does not
- Benefit considered beyond the operator
- Impact on affected groups recorded
- Environmental cost acknowledged
- Decisions documented, not assumed
Human-centred values and fairness
Rights, dignity, autonomy
- Human rights considered at design
- Discriminatory outcomes tested for
- Human autonomy preserved
- Redress that a person can reach
Transparency and explainability
Meaningful, not merely available
- Disclosure that AI is in use
- Explanation fit to the audience
- The basis of an outcome available
- Not defeated by a licence agreement
Robustness, security and safety
Over the whole life
- Function reliably under real conditions
- Traceability of data and decisions
- Risk managed continuously
- The ability to override or withdraw
Accountability
Someone answers
- Named accountability for outcomes
- Not delegated to a vendor
- A trail from finding to change
- Records that outlive the team
Why it matters commercially
The practical value
- One vocabulary across jurisdictions
- Maps onto AI RMF and ISO 42001
- Recognised without explanation
- A defensible starting position
What alignment has to mean
Anyone can say they follow a set of principles. Four things separate a claim from a position.
A written position
Each principle mapped to what you actually do, with the gaps named rather than smoothed over.
Evidence per principle
Test results, records and decisions — not a policy that restates the principle back.
Named owners
Accountability that survives a reorganisation, because it is attached to a role rather than a person.
A map to what binds you
The Principles are the foundation. The binding framework is the thing you have to satisfy.
From principles to evidence
-
Phase 1
Position
Say what you mean by alignment
- Inventory AI systems in scope
- Map each principle to current practice
- Name the gaps honestly
- Identify the binding frameworks
-
Phase 2
Evidence
Make the claim checkable
- Testing against fairness and robustness
- Disclosure reviewed for meaning
- Ownership assigned per system
- Records generated by the process
-
Phase 3
Map across
One body of evidence
- Tag findings to AI RMF categories
- Tag findings to ISO 42001 clauses
- Tag findings to EU AI Act articles
- One assessment, several answers
-
Phase 4
Maintain
Keep it true
- Re-assess on a fixed cadence
- Update when systems change
- Report to whoever is accountable
- Retire claims you can no longer support
Frequently asked questions
Are the OECD AI Principles binding?
Not on you. They are an intergovernmental recommendation that adhering governments commit to reflect in national policy. Where they bite is indirectly — through the national frameworks, regulator guidance and procurement terms that were drafted from them.
Can we be certified against them?
No. There is no certification scheme and no conformity assessment. What you can do is state a position, evidence it, and have that evidence stand up — which is the useful version of the same thing.
Is this worth doing if the EU AI Act applies to us?
Build to the Act; it binds you and the Principles do not. Where this earns its place is explaining to a board, a customer or a regulator in another jurisdiction how the pieces fit together.
How do they relate to NIST AI RMF and ISO 42001?
They sit underneath both. AI RMF gives you a structure for the risk work; ISO 42001 gives you a certifiable management system; the Principles describe what both are for. Findings map cleanly between all three, which is why we tag them.
What does "explainability" actually require?
Explanation that means something to the person receiving it. A feature-importance chart satisfies a data scientist and nobody else. The test is whether the recipient could act on what they were told.
Ready to evidence OECD alignment?
A written position per principle, evidence behind each one, and a map to the frameworks that actually bind you.
This page is guidance on how we scope an assessment, not legal advice.