AI Transparency Notice

Required disclosures under Regulation (EU) 2024/1689 (the "AI Act"). Gater uses AI in two clearly-bounded ways: assisting human authors when generating exam questions, and assisting human reviewers when grading or flagging proctoring anomalies.

1. Risk classification

AI in education that determines access, evaluates outcomes, or monitors test-taking is classified as high-risk under Annex III §3 of the AI Act. We treat all our in-product AI as high-risk and apply the corresponding obligations.

2. Where AI is used

• Question generation — Gemini family models draft items from teacher-supplied topics; every item is reviewed and approved by two human authors before it enters the live pool.
• Question quality scoring — clarity, bias, factual-confidence heuristics surface to the reviewer; the reviewer decides.
• Item-level psychometrics — IRT 3PL calibration runs on real attempts, not AI-generated guesses; recalibrates every 25 attempts.
• Grading assistance — for closed-form MCQ items, scoring is deterministic. For any AI-assisted scoring, the result is marked "provisional" until a human reviewer signs off.
• Proctoring anomaly flags — heuristic + AI signals create a risk tier; flags never auto-disqualify; a human always reviews.

3. Human oversight (Art.14)

No exam result, certificate, admission decision, or disqualification is issued by AI alone. A qualified human (university administrator or proctor) reviews and confirms. Candidates can always request a human re-review of any flag or score; the institution's appeal process applies.

4. Your rights under GDPR Article 22

Because every consequential decision involves a human, Art.22 prohibition on solely-automated decisions does not apply. If an institution opts in to fully automated grading in the future, candidates will be given a separate explicit consent and a prominent "Request human review" control.

5. Limitations and known risks

• AI-generated questions may reproduce subtle biases from training data; our review process and bias scorecard mitigate but do not eliminate this.
• Face matching can produce false positives or negatives, especially across demographics; mitigations include manual fallback and multi-modal liveness.
• AI confidence scores are not legal certainty.

6. Conformity & technical documentation

Technical documentation under Annex IV is maintained internally and provided to supervisory authorities on request. Conformity self-assessment (Art.43 internal control route) will be signed and registered before the 2 August 2026 deadline applicable to high-risk AI systems.

7. Contact

Questions about Gater's AI usage: dpo@gater.app.