EU AI Act Transparency

Skillvora may be classified as a high-risk AI system under Annex III(3) of the EU AI Act when used to evaluate learning outcomes that affect a person's access to or progress in education. We design the platform on that assumption and publish this transparency notice accordingly.

Human oversight (Art. 14): every Evidence Packet is faculty-reviewable before it reaches a gradebook. Faculty can override any rubric score, and rubric scoring uses transparent, faculty-authored criteria rather than opaque end-to-end models. The platform is built so that no consequential grade is finalized without a human decision unless the institution explicitly configures auto-passback.

Accuracy, robustness, cybersecurity (Art. 15): we publish rubric-score agreement metrics with human raters on representative scenarios in our model card, run adversarial-prompt regression tests on every model upgrade, and isolate student PII from model training. Skillvora does not use student transcripts to train foundation models.

Risk management & data governance (Art. 9–10): documented risk register, scenario-level bias testing across demographic dimensions where institutional data permits, and a published model card available to procurement on request.

Transparency to users (Art. 52): students are informed at session start that they are interacting with an AI system. AI-generated content (coaching notes, rubric reasoning) is labeled as such in the Evidence Packet.

Conformity documentation, including the technical file required under Art. 11, is available to institutional customers under NDA. Email aiact@skillvora.io.