1. Scope and control
AI may assist with generation, explanation, translation, speech, transcription, search, formative feedback and scoring suggestions. The School selects enabled capabilities, roles and classes. AI scoring is enabled by default for new school classes once the required compliance profile and current approval are present; the School can switch it off. Other School AI is disabled unless its required compliance profile and current approval are present.
2. Data rules
Only information necessary for the request may be sent to an allowlisted provider/model route. School requests require contractual no-training and shortest-retention/zero-data-retention settings. General usage analytics exclude prompt/response text. Automatic routing may not bypass the School’s agreed deployment or model policy. If TensorX is selected, there is no fallback to OpenRouter or another AI provider; unsupported features are unavailable. Requested custom installations use the Hetzner, Clerk and TensorX core stack described in the Subprocessor Schedule. A provider is inactive until its contract and downstream model endpoint are approved.
3. Transparency and AI literacy
Users are told when a feature generates or evaluates with AI. Documentation states intended purpose, input, provider category, known limitations, retention, human review and complaint route. Ekadem supplies administrators and staff with information to understand enabled AI; the School trains people for its context and teaches students how to question output.
4. Meaningful human oversight
AI output is a suggestion. A qualified person must have time, authority, competence and relevant source material to review it; understand limitations; disregard or override it; correct records; and answer a challenge. AI may not be the sole basis for admission, placement, final grades, discipline, cheating sanctions, special-education eligibility or denial of access/opportunity. A non-AI/manual path must be available where required.
5. Progress indicators and disclosure limits
For new classes, scoring is enabled by default after the compliance gate, but the School can switch it off. Ekadem evaluates turned-in work into a progress indicator for teaching staff. This is profiling under data protection law and is disclosed as such in the public AI notice and privacy notice. The output is probabilistic and may be inaccurate, incomplete, biased or misleading, including because a submission is ambiguous, multilingual or inaccessible; it must not be treated as a grade or relied on without review of the underlying work. Ekadem commits, as a product property and not merely a setting, that the indicator: is not a grade and does not become one; is disclosed only to teachers of that class and school staff with teacher access; is withheld from the student it describes and from their classmates, in the API response and in realtime delivery alike; is not written to a report, transcript or record and cannot be exported; and is never used to group, rank, level, advance, flag or refer a student. Cohort-level AI summaries follow the same disclosure limit. Feedback authored for a student is a distinct, labeled feature delivered only to that student. Changing any of these properties is a material new AI purpose under section 7.
6. Prohibited uses
- emotion recognition in education;
- social scoring or manipulative exploitation of a child’s vulnerability;
- unlawful biometric identification/categorization or sensitive-trait inference;
- general-purpose model training, targeted advertising or unrelated profiles from School Data;
- high-impact automated decisions without valid classification, impact assessment, law, notice, human review and appeal; or
- upload of specially protected records to a feature not expressly approved for them.
7. Risk and quality management
Ekadem documents intended use, provider/model version, material evaluation, security controls, complaints and incidents; tests accuracy, bias, accessibility and other quality risks appropriate to the use; monitors changes; and supports a DPIA/AI impact assessment. Because scoring is enabled by default for new school classes, the School must ensure staff know it is AI, review the underlying work, correct errors and provide a meaningful alternative when it is switched off or unsuitable. A material new AI purpose requires assessment and School approval before processing existing Student Data.
8. European AI Act readiness
Prohibited practices are not enabled. AI literacy applies to staff and deployers. Before any education use that may qualify as high-risk under the EU AI Act, the parties must classify it, complete required risk/quality/data/record/transparency/human-oversight/conformity and deployer assessments, and confirm the applicable implementation date. The existence of this policy is not a high-risk conformity declaration.