EU AI Act hiring compliance in 2026 costs less when the shortlisting runs on your own hardware: keep the model, the candidate data and the decision record on the machine, and the transparency disclosure and fairness note are generated as you hire rather than bought in. That is the saving. You stop paying for the Workday and SuccessFactors AI modules, you stop commissioning a separate bias audit and outside counsel for every campaign, and the evidence the regulator wants is produced by the system you own instead of assembled after the fact.
Why EU AI Act hiring compliance is a live cost in 2026
The rules stopped being a future problem this month. Industry reporting on the EU AI Act puts the transparency obligations into force on 2 August 2026 and now enforceable, with the fuller high risk duties for recruitment and candidate selection following from 2 December 2027. The Act treats systems that screen applications and rank candidates as high risk, which means an employer that uses AI to shortlist owes candidates a disclosure that AI was involved, has to keep logs of how the system reached its result, and has to be able to show the tool was tested for discriminatory impact. None of that is optional, and the clock on the hiring specific duties is already running.
The reason employers are tooling up now rather than in 2027 is that the evidence has to exist for the hires made under the new regime, and you cannot reconstruct a fairness assessment for a decision the system has already forgotten. The transparency piece bites first, the high risk file follows, and both need a record that ties a specific shortlist to the checks that were run on it.
What hiring AI compliance costs you today
Three bills stack up, and only one of them is the software. The first is the licence: the AI recruiting modules bolted onto Workday and SuccessFactors are priced per seat or per employee, so the meter runs on headcount whether or not you are actively hiring that month. The second is the audit: to satisfy an adverse impact or bias requirement, firms buy a separate testing tool or pay a vendor to run the four fifths style analysis, often per campaign. The third is counsel: the transparency disclosures and the fundamental rights framing get drafted by an outside employment lawyer, billed by the hour, and redrafted each time a role or a jurisdiction changes.
Underneath the invoices sits the exposure. Every candidate CV scored in a vendor cloud is personal data leaving your control, and for an EU facing employer that can mean a cross border transfer to reason about on top of the AI Act file. A shortlist you cannot explain is not a discount, it is a fine risk and a reputational one, and the accountability stays with the employer, not the software vendor.
How Ergon shortlists candidates on your own hardware
Ergon is our people studio, and a studio is a ready made application for one business function that runs inside a single system. Ergon builds AI Act aware hiring shortlists, answers HR policy questions and runs workforce analysis, and it does all of it offline on the employer's own hardware. It is built to replace the Workday and SuccessFactors AI modules, but the difference that matters is where the candidate data sits: on the machine, never in a vendor cloud.
The Assistant that drives Ergon runs on the company's own brain, a model built on the company's own roles, rubrics and hiring history rather than a shared cloud service. A shortlist comes with a fairness note attached, a record of the checks run against the ranking, and a transparency disclosure a candidate can be shown. Because the model, the applications and the output all stay on infrastructure the employer controls, the personal data being assessed never leaves the building, and the cross border transfer question largely goes away.
The fairness note and transparency disclosure, generated not commissioned
This is where the money actually moves. Under the AI Act a high risk hiring system has to be documented: a disclosure to the candidate, logs of the decision, and evidence that the tool was checked for discriminatory outcomes. Ergon produces that as a by product of making the shortlist. The fairness note records the checks run on the ranking, the transparency disclosure states that AI was used and how, and every run is sealed under post-quantum cryptography into the Open Audit Record, a signed, tamper evident log of who ran what, on which cohort, and when.
That does not hand you a certificate, and we are careful not to claim one. What it does is generate the evidence that supports the examination, on your own hardware, instead of paying a bias audit vendor and an outside lawyer to assemble it after each round. The record is local and verifiable, so when a works council, a regulator or a rejected candidate asks how a decision was reached, the answer is already on file.
What you replace, and what you save
| What you run today | What it costs you | With Mickai |
| --- | --- | --- |
| Workday AI recruiting module for candidate ranking | Per seat licence, and CVs scored in a vendor cloud | Ergon shortlists offline, no per seat meter and no data egress |
| SAP SuccessFactors recruiting AI | Priced per employee, running on headcount not on hiring | Ranking runs on owned hardware, the per head cloud fee disappears |
| A separate bias or adverse impact audit tool | Per campaign or annual subscription for the testing | The fairness note is produced with each shortlist, on device |
| Outside counsel for AI Act transparency disclosures | Hourly legal fees, redrafted per role and jurisdiction | The transparency disclosure is generated as a by product of the hire |
| Candidate data sent to a cloud service for scoring | Cloud processing plus a cross border transfer to reason about | Applications never leave the machine, each run sealed to the Open Audit Record |
How a hiring round runs end to end, on device
In practice a round moves through Ergon like this, and every step stays on the employer's own hardware:
- Load the role, the rubric and the applications into Ergon on the employer's own hardware, with no upload to any external cloud.
- The Assistant, running on the company's own brain, ranks candidates against the role's stated criteria and produces a shortlist.
- A fairness note is attached to the shortlist, recording the checks run against the ranking so a recruiter can see them rather than take them on trust.
- A transparency disclosure is generated for candidates, and a recruiter reviews and approves; the model shortlists but does not reject, send or complete anything irreversible on its own.
- Each run is sealed under post-quantum cryptography into the Open Audit Record, so the disclosure, the fairness checks and the decision are answerable later.
What specifically leaves the software budget
Fold it up and three recurring lines come off. The per seat or per employee AI recruiting module stops, because the owned system carries no seat meter. The separate bias audit fee stops, because the fairness note is produced with every shortlist rather than bought per campaign. And the outside counsel hours for transparency disclosures shrink, because the disclosure is generated by the system as it hires. What replaces them is a one off owned capability, paid for once and kept, on infrastructure the employer controls.
Frequently asked questions
When do the EU AI Act's hiring rules apply?
Industry reporting puts the transparency obligations in force from 2 August 2026 and now enforceable, with the fuller high risk duties for recruitment and candidate selection following from 2 December 2027. Because the evidence has to exist for hires made under the regime, employers are building the capability now rather than waiting for the later deadline.
What tools does Ergon replace?
Ergon is built to replace the Workday and SuccessFactors AI recruiting modules, plus the separate bias audit tooling and the outside counsel hours that hiring AI compliance currently pulls in. The saving is that the per seat licence and the per campaign audit fee both disappear, and the transparency disclosure and fairness note are generated on hardware the employer owns.
Does candidate data leave the building?
No. Ergon runs offline on the employer's own hardware and the Assistant runs on the company's own brain, a model built on the company's own data. The applications, the ranking and the output stay on infrastructure the employer controls, so the personal data never enters a vendor cloud or an external training set, which also settles the cross border transfer question.
Does Ergon make the hiring decision on its own?
No. Ergon shortlists and attaches a fairness note and a transparency disclosure, but a recruiter reviews and approves. The model does not reject a candidate or complete an irreversible action by itself, and every run is sealed under post-quantum cryptography into the Open Audit Record, so the decision and the checks behind it are answerable later.
Top comments (0)