The position, stated plainly
Article 22 of the GDPR restricts solely automated decisions that produce legal or similarly significant effects on a data subject. humaniser.eu returns a verdict — human, AI, or manual review — together with unsigned scan details identifying the model, the calibration card, and the per-sentence rationale. The verdict is intended as input to a human reviewer's decision. It is not the decision.
Customers integrating the API into a workflow that produces a legal or similarly significant effect — disciplinary panel, hiring rejection, retraction, refusal to publish — are responsible for the human-review step. The review record is built to be the artefact that a human reviewer reads. Critics dispute whether decision-aid framing is sufficient under Article 22 when the receiver in practice rubber-stamps the verdict; they are not entirely wrong, which is why the scan details surface the abstain band and the false-positive rate rather than a single number.
Right to explanation
Data subjects who receive an adverse decision in which a humaniser.eu verdict was used as decision-aid have the right to a meaningful explanation. The scan details, the applicable published evaluation card, and the per-sentence evidence are designed to support that explanation. The operator — the customer running the workflow — is the controller responsible for furnishing it.
Contact
Data subjects, regulators, and DPOs may contact info@humaniser.eu for scan-details verification, model-SHA confirmation, or calibration-card reproduction questions.