Legal
DPIA Summary
Public summary of NEB RADAR's Data Protection Impact Assessment for the closed beta. Last updated: 13 November 2026. The full assessment is available to mediator organisations and public partners on request (info@neb-radar.eu).
Scope
This assessment covers: (a) group sessions in the Mediator Space, including the planned sessions with 13 to 17-year-olds; (b) the AI companion and the Agent Advisory Council, and AI help in assessments (sentences found in the sources, open-data proposals); (c) usage statistics; (d) place photos added by organisations and by participants aged 18 or over; (e) beta access requests and accounts of invited participants. NEB RADAR publishes no rankings of people, groups or organisations and no individual answers.
Risks & mitigations
| Risk | Severity | Mitigation |
|---|---|---|
| A young person could be identified from what a group says about a place | High | Participants join a session with a code, without an account, a name or an e-mail address; only the hash of a random browser key is stored. Individual answers are never published and never shown one by one, not even to the mediator: only the group's pattern. What is published is the statement the group confirms together, with a group card (age band, size range, organisation, activity, season, place). Comments and notes are deleted 30 days after the session closes and the key link is cut. |
| Free text could contain personal details or a disclosure that needs help | High | Before any text reaches the AI, a deterministic safeguarding check runs; on a match no model is called, a fixed help text is shown and only the fact and time are recorded. Personal details in the text are masked before it leaves NEB RADAR. Comments are screened at submission as well. |
| Sessions with 13 to 17-year-olds carry obligations under Art. 8 GDPR and child-safety rules | High | Sessions with 13 to 17-year-olds are not open yet. They open only through mediator organisations, with consent through the organisation and, where the law requires it, the parents, after the safeguarding steps are complete. Protection sits with the platform, not with the mediators. |
| An AI answer could be wrong or read as an official or legal statement | Medium | The companion answers only from NEB RADAR's own fact sheet and records, with a deterministic check that blocks claims of official status, invented names or codes and scores the platform does not publish; otherwise it says it does not have the answer. The Agent Advisory Council answers only from the regulation register and carries a not-legal-advice line. Every AI surface is labelled as AI; nothing of a conversation is stored or used for training. |
| A place photo could show a person or reveal where young people meet | Medium | Photos must show no people; metadata (location, camera, time) is removed; an AI check looks for people and personal details. Photos from organisations are reviewed by a person at NEB RADAR before they appear outside the organisation. Photos from participants aged 18 or over carry a Creative Commons licence and a credit; the AI agent shows a photo at once only when it clearly shows a place without people or personal details, any other waits for a person, and every agent decision can be taken back. Under-18s cannot add photos. Suggestions by 13 to 17-year-olds are coarsened before storage. |
| An observation or a guided assessment by a Young Citizen could carry personal details or be read as a judgement of the person | Medium | Only signed-in participants aged 18 or over can write them. Personal details are masked before storage and before any AI call; the safeguarding stop runs first. Observations reach professionals as quotes without a name and can be removed by the author at any time. The evidence review agent rates evidence quality per measure only, never changes a class and never assesses the person; a professional decides. The learning path counts observations per measure and never ranks people; there are no points and no leaderboard. |
| An AI quote could be read as an answer or as a judgement of the project | Medium | In an assessment the AI only shows sentences from the sources stored with that assessment; it never chooses an answer, and the person chooses. A sentence is shown only when it is found word for word in the sources, two of three runs gave it and three checks found it directly about the question. Nothing is kept unless the person keeps it as an evidence note. The AI is never asked for numbers, measured bands or open-data indicators, and never about people. Open-data values (walking network) are proposals the person confirms. |
| Usage statistics could track individuals | Low | Recording starts only after consent; no name, e-mail, account, IP address or user agent string is recorded; a visit ID expires after 30 minutes; known bots are ignored; records are deleted after 12 months; consent can be withdrawn at the bottom of every page. |
| A processor outside the EU or a change of processor | Low | Hosting, database, authentication and files on Scaleway (Amsterdam); AI on Mistral AI (France) and, for pictures, Black Forest Labs (Germany), which may use design texts (personal details removed) to improve its models; e-mail through IONOS (Germany). No processor outside the European Economic Area. Sign-in with Google is optional and works without Google. The processor list and the privacy notice are updated before a change. |
Residual risk
After these measures the residual risk for the processing that is open today (adults, 18 to 30) is assessed as Low. Sessions with 13 to 17-year-olds stay closed until the safeguarding steps are complete and this assessment has been reviewed for them.
Review cadence
This assessment is reviewed before sessions with 13 to 17-year-olds open, when a processor changes, and at least every 12 months. A weekly automated check watches changes in the rules on safeguarding, data protection and AI; a person decides on every adaptation.