Article 64.4.d of the Spanish Workers' Statute, and equivalent duties elsewhere in the EU
Article 64.4.d of the Spanish Workers' Statute, introduced by Law 12/2021, gives the works council the right to be informed of
The right is not limited to platform work, it expressly covers access to employment, meaning recruitment, and it names profiling explicitly. It is not subject to the trade-secret balancing that applies in some other contexts.
Article 26(7) of Regulation (EU) 2024/1689 imposes a parallel duty across the Union: before putting a high-risk AI system into service at the workplace, employers shall inform workers' representatives and the affected workers that they will be subject to its use.
The company [TO COMPLETE: legal name] uses the HRBLADE platform, provided by AMISCON GLOBAL S.L. (NIF B70862099, Valencia, Spain), in its recruitment.
[TO COMPLETE: tick what applies]
Functions not ticked are switched off in the company's configuration.
Not processed as structured fields: sex, date of birth, age, nationality, ethnic origin, religion, disability or marital status. These are not requested from candidates.
Not performed: facial analysis, analysis of voice characteristics, or inference of emotions. CV photographs are not sent to any model. Audio is converted to text and only the text is assessed.
Each interview answer is assessed on three axes, 0 to 100:
| Parameter | What it measures |
|---|---|
| Relevance | The extent to which the answer addresses the question asked |
| Depth | The level of specific, substantive detail provided |
| Clarity | Structure and comprehensibility of the response |
A CV is assessed against the published requirements of the vacancy on three axes: match of experience, of skills and of education.
The application score is the average of the individual scores.
The system is expressly instructed to disregard the candidate's name, sex, age, origin and any gaps in employment history, and to assess only the content of their answers and their fit against the published requirements.
In voice interviews the system is additionally prohibited from drawing any conclusion about emotions, mood, confidence or personality traits from the manner of speaking.
[TO COMPLETE by the employer]
| Stage | Advance threshold | Rejection proposal threshold |
|---|---|---|
| CV screening | ||
| Interview |
If no threshold is configured, the score only ranks and highlights candidates and triggers no automatic action.
The hiring decision is made by a person. The score is advisory.
Where a score falls below the configured threshold, the platform proposes a rejection which appears in a review queue. A person confirms or dismisses it, and the action is recorded with their identity, the date and the time they spent on the decision.
Fully automated rejection, with no human involvement, is switched off by default. [TO COMPLETE: state whether the employer has enabled it. If so, the legal basis under Article 22 GDPR must be documented, along with the route to human review offered to candidates.]
Every AI-influenced decision is recorded with: type of decision, the function and model that produced it, score, applicable threshold, rule triggered, outcome, and whether a person reviewed it and who. Records are kept for 12 months and can be exported per candidate and per vacancy.
Candidates are informed, before the interview and when they apply, that AI is used, what it does, and that a person makes the decision. They can request human review of the decision, an explanation of how their answers were assessed including what would have had to differ for a different outcome, and access, rectification or erasure of their data, through a page that does not require creating an account.
The provider runs a controlled-comparison test: profiles identical in qualifications that differ in exactly one signal at a time (forename as a proxy for sex and origin, graduation year as an age marker, gaps in employment, non-native phrasing). Each variant runs through the real scoring path and the impact ratio is measured against the four-fifths rule, along with consistency between matched pairs.
The test runs on a schedule and again whenever the model, a scoring instruction or a threshold changes. Findings are remediated before the change is released. The full methodology and a summary of the most recent run are available through the employer.
Data is stored in the European Union, in Amsterdam, the Netherlands. The transcription engine runs on the provider's own infrastructure in the same location. Text sent for analysis is processed by the AI model provider; where that processing occurs outside the European Economic Area, the transfer is covered by Standard Contractual Clauses together with a transfer impact assessment.
Changes of model and of scoring logic are recorded and dated. The employer receives at least 30 days' notice of any change that may materially affect scoring behaviour.
Public AI Transparency Statement: https://hrblade.com/ai-transparency/
The employer may request from the provider: instructions for use, the sub-processor list, an extract of the record of processing activities, and the bias audit summary.
Company contact for this document: [TO COMPLETE: name, position, email]
Date provided to the works council: [TO COMPLETE]
