AI Usage Policy
How Bonjour uses AI: Sanctuary and Open modes, user responsibilities and compliance with the French AI Law (August 2, 2026) and GDPR.
Last updated: July 20, 2026 · Version 1.0
1. Our principles
The association "Un Pour Tous Et Tous Pour Un" uses AI strictly as an assistance tool for its members. AI makes no binding decisions and never replaces professional advice.
2. Operating modes
Sanctuary Mode (Zero-Knowledge) — Your data is encrypted on your device (AES-GCM 256). The association cannot read it. Files are automatically deleted after 48 hours. Encryption keys never leave your browser. Recommended for maximum privacy.
Open Mode — Data is processed on our servers for extended features and speed. You explicitly consent to this processing and can switch back to Sanctuary at any time.
3. AI limitations
L'IA fournit des recommandations, aide à l'analyse et génère des modèles.
AI does not make binding decisions, does not replace professional legal, medical or financial advice, and guarantees no outcome.
4. Your responsibility
By using AI tools, you agree to make final decisions yourself, to verify important outputs (legal, financial, medical) with a qualified professional, and not to hold the association liable for those decisions.
5. Compliance
Our services comply with GDPR (EU Regulation 2016/679), the French AI Law (August 2, 2026) and UNESCO's ethical AI principles.
6. Your rights
Choose your mode (Sanctuary or Open) at any time. Get an explanation of any AI recommendation. Request human review. Delete all your data. Leave the association at any time.
7. Contact
AI officer: Sergueï Chevtchenko. For any question: contact@unpourtousettouspourun-asso.fr
8. Register of AI systems
In accordance with Article 6 of the French AI Law, we publish the list of AI systems in production, their purpose and risk category.
| System | Provider | Purpose | Risk |
|---|---|---|---|
| Gemini 3 Flash | Google via passerelle IA | Conversational assistant, summaries, drafting help. | Limited |
| Gemini 3 Flash (Analytics) | Google via passerelle IA | Natural-language analytics queries over aggregated data. | Limited |
| Smart Import Extractor | Google via passerelle IA | Data extraction from imported files (CSV, PDF). | Limited |
9. No use of your data for model training
We formally declare that member data — messages, files, prompts, content from Sanctuary or Open mode — is NEVER used to train, retrain or fine-tune AI models, neither by the association nor by our providers (Google via AI gateway) under their API terms.
No personal data is sold, transferred or shared with third parties for model training.
10. Data Protection Impact Assessment (DPIA)
A Data Protection Impact Assessment (DPIA) has been carried out for each AI system listed in §8. Public summary:
• Legitimate purpose (member assistance) and proportionality verified. • Minimization: Sanctuary mode encrypts inputs client-side; files auto-delete after 48h. • Legal basis: explicit consent and association's legitimate interest. • Residual risks: AI hallucinations (mitigated by warnings and the right to human review).
The full report is available upon request from the AI officer (Sergueï Chevtchenko) at contact@unpourtousettouspourun-asso.fr.
11. AI Ethics Committee
An AI Ethics Committee oversees AI usage within the association. It is chaired by the AI officer, Sergueï Chevtchenko, and meets at least twice per year to review new systems, incidents and complaints.
To reach the committee: contact@unpourtousettouspourun-asso.fr (subject: "AI Ethics Committee").
12. AI incident register
We maintain an internal register of AI-related incidents (bias, errors, malfunctions, privacy breaches). Serious incidents are notified to the CNIL and to the affected persons within legal deadlines.
As of today: no serious incident reported. The public summary is updated at least once a year.