Secours Islamique France AI strategy: sovereign architecture, security and governance

Location

10 rue Galvany

City

91300 MASSY

Other required skills (specific tools, coding, manual skills, etc.)

Digital strategy & IT

Constraints & risks (logistical, physical, emotional)

Stack hosting the Agentic AI is on-premise, with no internet access or PowerShell / Sensitive data: no access to production systems. Work will be carried out on anonymised documents. Prior validation by the IT/security team (Samir) is required for any network access.

Secours Islamique France AI strategy: sovereign architecture, security and governance

Description

Secours Islamique France (SIF) has been engaged since 2024 in an ambitious digital transformation integrating artificial intelligence at the service of its humanitarian mission. Against a backdrop of rapid growth in AI use and tightening regulation (GDPR, EU AI Act), SIF has set up an AI Steering Committee (Copil AAI) tasked with validating and steering strategic orientations. The project involves mobilising the expertise of Wavestone consultants to confirm and refine two structuring pillars of SIF's AI roadmap: 1. SOVEREIGN AI ARCHITECTURE: Validation of the choice of an on-premise stack (SIF server, Ollama, LiteLLM, Docker) for processing sensitive data, arbitration on the open-source models deployed (Deepseek-R1:7b for reasoning, Qwen2.5-Coder:7b for execution), and consistency of the overall architecture (n8n, FastAPI, AI agents, M365/SharePoint integration). 2. CYBERSECURITY & COMPLIANCE: Review of AI risks, validation of the governance framework (AI Charter, DPIA, ISO 27001/42001), security posture of the local infrastructure (VLAN, Samir IT access), and recommendations on attack vectors specific to LLMs (prompt injection, data leakage). The aim is to obtain independent external validation of SIF's AI strategy before rollout to production, while upholding data sovereignty and the regulatory constraints specific to the humanitarian sector.