Platform · Security
GDPR and AI Act compliance by design
Data hosted in the European Union, optional 100% EU AI processing, exportable audit log. Edugent is compatible with Claude, GPT, Mistral and its own instances hosted by OVH in France. Articles 14 and 50 of the AI Act are implemented in the code.
Data hosted in the EU
Database and documents on Supabase (Ireland, European Union). Optional 100% EU AI processing, knowledge base included, if you require maximum sovereignty. Full list of sub-processors in the legal notice.
Encryption in transit and at rest
TLS 1.3 in transit, AES-256 at rest.
Everything is logged
Module runs, AI model calls, kill switches and record views, in a single timeline of metadata. Kept for one year at most, CSV export by your administrator.
AI Act
Compliant today, ahead of the AI Act deadline
AI systems in education are classified as “high-risk” by Annex III of the AI Act. Date of application: 2 December 2027 (following Regulation (EU) 2026/1744 published on 24 July 2026).
AI Act · Article 14
Human oversight
Decisions always remain human: the twin writes drafts, an identified human approves, edits or rejects them, and every run can be audited after the fact. The twin principles embody this requirement by design.
AI Act · Article 50
Transparency
By default, a transparency line at the foot of every message prepared by a twin, and a “Write to [name] in person” button on the twin's public page. Recipients know that an AI assistant took part in the reply and can reach a human without friction.
AI Act · Annex III
High risk
Systems used to decide on access, admission or grading in education are high-risk. At Edugent, AI prepares and proposes. The decision remains human.
The digital twin principles are our by-design implementation of the requirements of AI Act Articles 14 and 50.
Read the principles →Security FAQ
Questions from CIOs and DPOs
Talk to our security team
For a joint audit, a tailored DPA or a technical Q&A session with your CIO: we'll arrange it.