The claim
IEEE calls CertifAIEd “the global mark of ethical and trustworthy AI” (IEEE, vendor). On 13 November IEEE Spectrum ran a piece by IEEE’s own microcredentials team arguing that engineers need credentials to build trustworthy AI (IEEE Spectrum, vendor). The implied promise to a buyer is simple: a system or a person with the mark can be trusted.
The scheme is real and structured. The promise is wider than what it certifies.
What the mark rests on
CertifAIEd has two separate certifications, and they certify different things (IEEE, vendor; IEEE Europe, vendor):
| Certification | What is assessed | What you receive |
|---|---|---|
| Product | Ethics governance during the AI system’s design, against criteria picked after an ethics risk profile, across four sets: transparency, accountability, algorithmic bias, and privacy | A certificate for the product that was assessed, and the right to use the mark |
| Professional | Whether a person can apply IEEE’s AI ethics framework, after training and an exam; eligibility is one year or more of experience using AI | A personal credential |
Three things are not on either page: a price (“size and scope dependent”), how long a certificate stays valid, and whether a new model version needs a new assessment (IEEE Europe, vendor).
The one public case
The case IEEE points to is the City of Vienna. Its write-up, by a data scientist at Wiener Stadtwerke, describes the assessment of an email classification system that sorts customer requests at Wien Energie (IEEE, vendor):
- A panel of five IEEE experts rated how likely the system was to undermine each of 26 ethical values, and put it in a low-risk class.
- From the four criteria sets, the panel selected one: accountability. IEEE supplied 43 accountability criteria.
- Wiener Stadtwerke wrote a 150-page “Case for Ethics” with evidence for each criterion. IEEE’s report confirmed it justified certification.
That is a serious piece of work. It is also an assessment of one system, on one of four criteria sets, written up by the organisation that was assessed. The report itself is not published. The OECD catalogue lists the scheme as “implemented in multiple projects” and names Vienna as its one use case; it gives no count of certified products (OECD, independent).
What it proves, and what it does not
| The mark is evidence that | The mark is not evidence that |
|---|---|
| An outside panel reviewed the product’s design-time governance against named criteria | The model is accurate. Accuracy is not one of the four criteria sets |
| The vendor wrote down, and evidenced, who is accountable for outcomes (where accountability was selected) | Transparency, bias, or privacy were assessed. Only the selected sets were |
| The assessed product met those criteria at the time | Today’s version still does. No validity period or retest rule is published |
| A person can apply IEEE’s framework (professional credential) | That person’s product is certified |
| The vendor took part in a voluntary scheme IEEE says aligns with the EU AI Act | Any legal obligation is met. Nothing on the pages we read says so |
The result: the mark certifies a process on a chosen slice of four criteria sets, for one product, at one point. Read it as that and it is useful. Read it as “trustworthy AI” and it is not.
Where this breaks
- It is still more than most vendors offer. A 43-criterion assessment by an outside panel, with a written evidence file, beats a slide that says “responsible AI”. Do not discount the mark to zero.
- Labels change what buyers believe. In a study of 12 interviews and a 302-person census-representative survey, certification labels significantly increased end users’ trust and willingness to use AI. Labels eased data concerns such as privacy; performance concerns were harder for labels to address (Scharowski et al., independent). That is the reason to read the label closely, not a reason to ignore it.
- We read the scheme, not an assessment. The criteria texts, the Authorized Assessor Registry, and the Vienna report were not available to us. If IEEE publishes them, this note changes.
Five questions before you give the mark weight
- Step 1. Which product, which version, which date? The certificate is for the product assessed. Match it to what you are buying. Owner: procurement.
- Step 2. Which criteria sets were selected, and why? Ask for the ethics profile result. One set of four is a different claim from four of four. Owner: AI governance lead.
- Step 3. Who assessed it? Ask for the assessor and check IEEE’s Authorized Assessor Registry. Owner: vendor risk.
- Step 4. Can we see the evidence file? Ask for the Case for Ethics, or an extract under NDA, and the assessor’s feedback. Owner: AI governance lead.
- Step 5. What covers accuracy and the law? The mark does not. Ask for performance evidence on your own data and take compliance to counsel. Owner: business owner with legal.
Missing
- A price, a validity period, and a retest rule for product certification.
- The full criteria sets and how they map to the IEEE 7000 series.
- A public register of certified products, and how many exist.
- The Vienna assessment report.
- Independent evidence that certified products perform or behave better than uncertified ones.
Frequently asked questions
Does an IEEE CertifAIEd mark mean an AI system is accurate?
No. IEEE describes product certification as an assessment of ethics governance during AI system design, against criteria for transparency, accountability, algorithmic bias, and privacy. Accuracy is not one of the four criteria sets, so ask for performance evidence separately.
Does an AI ethics mark show EU AI Act compliance?
IEEE's page says product certification aligns with emerging requirements, including the EU AI Act. Nothing on the pages we read says the mark meets a legal obligation. Treat it as supporting evidence and take the legal question to counsel.
What is the difference between an IEEE CertifAIEd professional and a certified product?
Professional certification says a person can apply the IEEE AI ethics framework. Product certification says one assessed product met the selected criteria. A consultant holding the credential does not make their product certified.
Living page. Last updated 15 Nov 2026. We update this note when IEEE publishes validity, pricing, or a product register, or when we see a full assessment. Weighing an AI vendor’s governance claims? Talk to us.