Certifying the Human Behind the Algorithm
Certifying that a named human remains accountable for every decision that matters.
Somebody was refused something this morning, and nobody can tell them why.
Hiring. Lending. Insurance. Healthcare. Parole. These decisions are increasingly made, or heavily shaped, by automated systems — and the person on the receiving end usually never learns that a system was involved at all.
Technology can be audited. Code can be reviewed. But accountability is not a property of software; it is a property of a person. Right now there is no recognised standard for who that person is, what they must know, or what they owe the people their systems decide about.
AIC certifies that a named human remains accountable — and publishes the result so anyone can check it.
Which closes a door. When an automated decision causes harm, the position available today is that the system did it and nobody understood it well enough to be responsible — no individual knew, so no individual answers. A certified organisation cannot take that position. Someone has signed a declaration accepting personal accountability for those decisions, and was tested on whether they could describe how the system actually behaves without checking with anyone else.
That is a defence given up, not a badge acquired. It is also the reason the mark is worth anything.
A decline letter, in full
“This decision was made in accordance with our internal credit policy. We are unable to provide further detail regarding the specific factors involved. This is an automated notification. Please do not reply to this message.”
Four sentences. No reason, no person, no way to reply, no route to challenge it. We score letters like this against a seven-part rubric, and a score below 40 out of 70 blocks certification outright.
How would you score this letter?
Seven dimensions, ten seconds. Score it, then see what our assessor scored it and where you disagreed. This is requirement EM-1 of the published standard, run exactly as it runs in an assessment.
Dear Applicant,
Thank you for your recent application. After careful consideration, we regret to inform you that we are unable to approve your request at this time.
This decision was made in accordance with our internal credit policy. We are unable to provide further detail regarding the specific factors involved.
This is an automated notification. Please do not reply to this message.
Customer Operations
Score it yourself
0/7Seven dimensions from the rubric AIC uses on real adverse communications. Nothing you enter leaves your browser.
Five rights, forty-four tests
Principles that cannot be measured are decoration. Each right below resolves into specific requirements with defined evidence, and every one of them is published.
Human Agency
A named individual is accountable for the decision, and can actually intervene in it.
Explanation
The reason a person is given is the reason that actually operated. Anything else is rationalisation.
Empathy
Adverse decisions are communicated the way a person receives them, not the way a system emits them.
Correction
Getting a decision wrong is recoverable, and the record shows it demonstrably happens.
Truth
People know an AI is involved before it affects them, not afterwards in the terms.
Everything we assert is somewhere you can go and look
A certification body that asks to be trusted has already lost the argument. These are the four things you can verify about AIC without speaking to us.
The full standard — what each test demands and what evidence satisfies it.
Read the standardEach carrying its own verification date and a link to the primary source.
Open the mapThe register is empty because nobody has been assessed yet. We publish that.
See the registerAIC is not accredited by UKAS, SANAS or anyone else. The application status is public.
Where we standNot ready for a full audit? Start with AIC Aware.
Twenty questions, the same four categories AIC audits against, about ten minutes. You get a risk-level snapshot and a downloadable PDF — no account, no cost. It is self-declared, not verified, and we say so throughout: only an independent AIC audit earns the Certified mark and a place on the public registry.
Your regulator's language, not ours
AIC's requirements were written against the instruments that already apply to you, so an assessment produces evidence you can use elsewhere rather than a second vocabulary to maintain.
See where regulation stands, by country- ISO/IEC 42001
- AI management systems. The management-system standard for AI, and the scheme AIC is pursuing accreditation to certify against.
- POPIA §71
- Automated decisions. South Africa's binding hook: no solely automated decision with legal or substantial effect, without safeguards and an explanation.
- EU AI Act
- Risk-tiered obligations. Transparency duties already apply; high-risk obligations are deferred to December 2027, not cancelled.
- NIST AI RMF
- Risk management. The voluntary US reference framework, and the language most American procurement is written in.
The register is empty. We would rather it stayed that way than fill it badly.
AIC is assembling a small founding cohort — organisations that shape the methodology, are assessed against it first, and carry the mark before anyone else. Their audits are also the witnessed activity our accreditation depends on, so the arrangement is reciprocal rather than promotional.