Between 2019 and 2021 the OECD, UNESCO and IEEE each set out what a responsible AI system owes the people it affects. Six ideas recur in all three - fairness, transparency, accountability, privacy, safety and human oversight - and every AI law written since has borrowed all six of them.
AI ETHICS · GOVERNANCE AUDIT · ANNUAL
Your published AI ethics principles are not a programme.
Six pillars, one governance structure and the evidence that each one operates - read against UNESCO, the OECD and IEEE, once a year.
Fairness
Transparency
Accountability
Privacy
Human oversight
Our promise
“Values are a statement. Practice is evidence.”
Every finding is written against a clause of the instrument itself — defensible line by line, to anyone who asks. The fee is fixed at $6,500, and nothing is charged until you approve it.
- Each additional tool
- $1,500
- Re-audit, same scope
- $4,200
- Renewal, every twelve months
- $5,500 locked
This costs more than the estate ladder, and it should. The ladder is a private assessment written for you. A framework audit produces a published summary iDharma maintains for twelve months - a notice template where the law requires one, a 60-day expiry warning, and a quarterly check that the summary is still live and still linked.
Request this auditThe case file
AI ethics, in three chapters
Most organisations have published a version of those six and stopped there. The principles have no owner, block no decision, and have never once held a release. Meanwhile the systems shipped and the thresholds moved, and nobody wrote any of it down while it was happening.
Our work is to close that gap in writing. We test the six pillars against records you already hold, review the four structures meant to keep them running, and hand you findings, a charter with an owner against every principle, and a policy register - written so you can defend each line.
What is an AI ethics programme?
Principles that can stop a release.
Principles with a named owner, and a record of where they blocked something.
- Anyone whose systems shape decisions about people
- Teams selling AI into regulated or enterprise buyers
- Organisations that have published AI principles already
- Boards asked to state what their AI exposure actually is
- Before a system that touches people reaches production
- When a buyer or regulator asks you to evidence a principle
- Every twelve months a system stays in use
- Whenever a model, threshold or use case materially changes
Whose duty is it
The duty is yours. The model is theirs.
The organisation that deploys it
Every instrument in this field puts the duty on whoever puts the system in front of a person. Setting the principles, deciding who signs off, keeping the evidence and telling people how the system affects them are all yours. None of them can be discharged by the people who built the model, however good they are.
The people who built the model
Carries duties of its own under the EU AI Act and under contract, but not yours. Many providers publish genuinely good model and system cards. The difficulty is not their quality - it is that they describe the model, while the ethical question is about your particular use of it, in your own context.
When a published principle counts
A principle counts when it has a named owner, a decision it can block, and a record of it having blocked one. Published on a web page with no owner, no gate and no evidence, it is a statement of intent - and that is exactly how a regulator, a journalist or a procurement reviewer will read it.
What most teams assume
“We published principles, so we have a programme.”
What the instruments ask
A principle that stopped nothing is a statement.
It is the most common finding we write up.
- Who it is for
- Product & platform teams
- Boards & audit committees
- Risk & compliance
- Legal & privacy
- AI vendors under procurement review
Why this matters in 2026
02
The EU AI Act, the Colorado AI Act and NYC Local Law 144 all enforce the same six ideas. The principles are voluntary; their content is not.
03
An assessment ages. Models drift and thresholds get relaxed, so a finding from last year describes a system you are no longer running.
04
Your principles page is checkable from the outside. A fairness claim you cannot evidence is still a claim, and that is how a regulator reads it.
The 60-second check
Three questions. Then you’ll know.
No email. No signup. A starting point, not a finding.
Opens in a panel. Close it any time - your answers are kept.
AI ethics & governance
Your scope check
The calendar
Four moments, and not in the order you’d guess.
Three of these run once and then repeat; the third runs on every release. They overlap, and none of them can be started from the one after it.
-
Charter
Weeks 1-6Your published principles, the committee that owns them and the decisions it may stop - by name, not by role.
-
Standards
Weeks 7-14Policies and thresholds written down, so a team can tell before it builds what will fail review later, and why.
-
Assess
Before releaseEvery system in scope tested against the six pillars, with the evidence kept as it stood on the decision date itself.
-
Re-review
AnnuallyModels drift, and so do the people who signed them off. A twelve-month re-review against your own baseline.
The trap
Teams write the principles last, after the systems are already live, and the charter then has to describe what shipped rather than govern what ships. Everything downstream inherits that: a committee cannot decline a release it was never asked about, and an assessment cannot find evidence nobody was told to keep.
Requirement & coverage
What the instruments ask, what we ship
12 requirements, and the artefact that answers each one. Paired, so every claim on this page can be checked against the requirement beside it.
- Whose duty it is The deployer, not the model provider
- A written determination of who owns each system and each principle, so a vendor model card is never mistaken for your programme.
- What is actually in scope Systems that shape a decision about a person
- An inventory of AI systems with the decision each one touches - including the bought ones nobody logged as AI.
- Fairness Disparate outcomes, measured not asserted
- Outcome rates by group for every system that affects people, with the comparator and the threshold stated.
- Transparency People are told when AI shapes a decision
- A check of every disclosure you make against what the system actually does, and the gaps between the two.
- Accountability A named person who can stop a release
- The decision rights mapped to real names and roles, and the escalation route when a review says no.
- Privacy Lawful basis, minimisation, retention
- Data provenance and retention tested against your own notices - the pillar most often assumed to be covered.
- Safety and robustness What the system does when it is wrong
- Failure modes, degradation behaviour and the monitoring that would tell you before a customer does.
- Human oversight Review that can change the outcome
- Evidence that a human reviewer has overridden the model, with the rate - the difference between oversight and sign-off.
- Board oversight AI reaches the board as a standing item
- Minutes, terms of reference and the reporting line reviewed for whether AI risk actually arrives there.
- Ethics committee A body with membership, quorum and teeth
- Charter, membership and decision log reviewed - including how often the committee has said no.
- Policies and standards Written before the build, not after it
- A policy register mapped to the six pillars, with every gap named rather than left to be inferred.
- Monitoring and audit The programme checks itself between reviews
- The metrics, the thresholds and who reads them - plus what happened the last time one was breached.
The engagement
Your AI programme, independently read
From a model you trained to a feature you licensed.
-
Inventory
Which systems are in scope, and who currently decides they may ship.
-
Assess
Six pillars tested against evidence, not against your policy text.
-
Report and publish
You see the draft first. Then findings, the charter and the register - dated.
Struck in your favour
Why boards choose iDharma to read their programme
Genuinely independent
We build, resell and operate no AI systems, and we take no fee tied to what the assessment finds.
Evidence, not policy text
Every finding rests on a record you already hold, so a reviewer can check it without taking our word.
One engagement, end to end
Pillars, structures and policies sit in one scope, so no gap falls between two separate reviews.
Mapped to what binds you
Each pillar is tied to the instrument that makes it enforceable, so the work is never only advisory.
Four marks, struck on every report.
Deliverables
What you get
Concrete artefacts, each with a name and a format - you know what lands before you buy.
AI ethics audit report
The full assessment in one document: a finding against each of the six pillars - fairness, transparency, accountability, privacy, safety and human oversight - each mapped to the instrument that makes it enforceable, ranked by consequence to the people affected, and written in plain language rather than the standards'.
Fairness results
The outcome tables in a workbook, so your own analysts can reproduce every figure in the report, cell by cell, without asking us for it.
Ethics charter
Your principles rewritten with an owner and a decision gate against each, marked up ready to publish - as HTML and PDF.
Committee pack
Terms of reference, a membership and quorum model, the escalation route, and the decision log the committee is meant to leave behind.
System scope memo
Which systems were in scope, which were not, and why each call was made - the written record that stands behind your AI inventory.
Remediation shortlist
Where a pillar fails, what to look at first, and why. Findings ranked by consequence to the people affected, not by ease of fixing.
Policy register
Your existing policies mapped to the six pillars, with each gap named and sized - so the writing that follows is scoped, not guessed.
Format & fee
Real numbers, upfront.
- Scope
- Six pillars, four structures
- Evidence
- Records you already hold
- Re-review
- Every rolling twelve months - $5,500 against your known baseline
The pillar set is fixed, so the fee is flat - only the inventory varies, and we count it with you before we start.
Request this audit
AI ethics · Named engagement
$6,500 flat
- Findings across all six pillars
- Outcome rates by group, with comparators
- Ethics charter with owners against it
- Policy register and committee pack
Show your hand
Four things you have to be able to produce
An ethics programme is not graded on intent. Each of these is either in your hand on the day someone asks, or it is not.
The charter,
owned
Your principles with a named owner against each one, and the route a concern travels. A principle nobody owns cannot be breached, so nobody ever reports one.
The register,
current
Every AI system in use, what decision it touches and who signed it off. Most inventories we are shown are missing the systems that were bought in rather than built.
The decision,
logged
A record of a review that actually changed something - a release held, a feature cut, a model rejected. Oversight either leaves a trace or it never really ran.
The numbers,
by group
Outcome rates by group for every system that affects people, with the comparator stated and the date on the run. The table itself is the evidence, not the assurance.
Four cards, and the date on each one is part of the card.
FAQ
Plain answers
Binding, what we measure, and what wrong costs. Answered straight.
Request this auditAre AI ethics frameworks actually binding on us?
Not of their own force - UNESCO, the OECD and IEEE bind nobody directly. Their content is what became binding: the EU AI Act, sector regulators and your own contracts all borrowed from them.
We published our AI principles. Is that a programme?
Almost never - this is the most common finding we write up. A principle counts when it has a named owner, a decision it can block, and a record of having blocked one. Most published sets have none of the three.
What exactly gets measured?
Six pillars against evidence: fairness, transparency, accountability, privacy, safety and human oversight - plus the four governance structures behind them. Every finding rests on a record you already hold.
How long does an assessment take?
Typically three to five weeks from document hand-over for a first assessment, longer where the system inventory turns out to be bigger than expected - which it usually does. Scope is agreed with you before anything is charged.
What are the consequences of getting this wrong?
No fine attaches to the principles themselves - it arrives through the instruments that adopted them: the EU AI Act, sector regulators, consumer-protection law, contract termination, and a failed procurement review.
Get started
Request your ethics audit
Tell us what you run, and we come back with a scoping call within one business day.
What we need from you
Nothing you do not already have. Most of this is a folder someone can assemble in an afternoon, and we tell you exactly which documents we need before you commit.
- Which AI systems are in use, and what each one decides
- Your published principles, if any, and who wrote them
- Any review or sign-off records from the past year
- Outcome data for the systems that affect people
- What is driving the timing: a buyer, a regulator, a date
What happens next
- You send the five items we need.
- We call to scope it within one business day.
- Nothing is charged until you approve the scope.
Sources & standing
Where this page gets its facts
Where the claims on this page come from, and what they are worth - stated, not assumed.
What it is drawn from
- UNESCO Recommendation on the Ethics of AI
- OECD AI Principles · IEEE 7000 series
- OECD
- May 2019, revised 2024
- UNESCO
- November 2021
What it means
- None of these instruments binds a company of its own force — their content binds through the laws that adopted it.
- Where a fairness definition is genuinely contested, our reports show the result under each reading rather than pick one.
Scope & limitation
- General information, not legal advice, and no professional relationship arises from reading it.
- There is no accredited certification for AI ethics. This is an independent assessment, and it does not issue one.
Something on this page out of date?
Tell usFrom Insights
Before you commission one
What an AI Governance Framework Actually Contains
Five working parts, not a policy document. What each one has to do, how to tell whether yours is real, and why a framework is not the same thing as compliance.
Read the guide
INSIGHTS
Startups, Meet Your AI Stack: Budget‑Friendly Tools That Scale
For early-stage founders, building an AI-powered toolkit doesn’t have to break the bank. From ideation to growth mode, here’s how startups can tap into affordable, effective AI tools to autom
INSIGHTS
The Missing Link in Corporate AI Training
In today’s fast-paced AI era, companies are investing more in training—but why is most of it still missing the mark?
STANDARDS EXPLAINED
Colorado SB 26-189: Four Duties, and the One Nobody Budgets For
Notice before, disclosure after, human review on request, records for three years. Three are policy changes. The second is an engineering project, and it arrives late.
