Required skills
- Security Architecture
- Threat Modeling
- Risk Analysis
- EBIOS RM
- Multi-tenant Architecture
- Data Encryption
- Key Management
- KMS
- HSM
- SSO
- MFA
- Least Privilege
- Privileged Access Management
- Audit Logging
- GDPR Compliance
- AI Act
- NIS 2
- Penetration Testing
- Disaster Recovery Planning
- Business Continuity Planning
Languages: French
Job description
Seeking an expert in Architecture and Cybersecurity to ensure a legal AI product has a secure, GDPR-compliant, and European-hosted architecture. The role involves acting as a trusted architect and accelerator for the team, focusing on security by design, data protection, identity management, and resilience.
Project overview
Expert Architecture & Cybersecurity (Legal AI Product)
Mission Objective
Ensure that the AI product launched in pilot in January is based on a secure, GDPR-compliant, and European-hosted architecture, capable of reassuring our clients (including during their audits and security questionnaires) and withstanding rigorous examination.
The expert will act as both a trusted architect (framing, deciding, validating) and an accelerator for the team (transferring knowledge, providing tools, unblocking issues), without replacing them.
Missions and Responsibilities
- Secure Solution Architecture
- Review and challenge the target architecture of the product (application, AI / LLM components, databases, IT system integrations) using a security by design approach.
- Conduct a risk analysis (EBIOS RM type or equivalent) and threat modeling, including risks specific to AI: prompt injection, data leakage via model responses, source poisoning, misuse.
- Define security requirements and translate them into concrete, prioritized technical choices for the team.
- Data Segmentation and Protection
- Design strict data segmentation between clients (multi-tenant architecture) to prevent any data cross-contamination, including in search indexes, caches, and context provided to the model.
- Define the data encryption strategy in transit and at rest, and key management (KMS / HSM, rotation, role separation).
- Oversee data usage by AI providers: no reuse for training, retention periods, minimization, and pseudonymization where relevant.
- Identity Management and Traceability
- Define the authentication and authorization model (SSO, MFA, least privilege, privileged access management).
- Implement complete logging of accesses and actions (who accessed what, when), which is tamper-proof, usable, and can be provided to a client upon request.
- Contribute to security monitoring and incident management procedures.
- Sovereignty and Hosting in Europe
- Ensure all processing, including AI model calls, takes place within the European Union.
- Evaluate hosting providers and suppliers (cloud, LLM, subcontractors): data location, exposure to extraterritorial laws, certifications (ISO 27001, SOC 2, SecNumCloud if applicable).
- GDPR and Regulatory Compliance
- Work with the DPO to produce the impact assessment (PIA), update the record of processing activities, and frame subcontracting agreements (DPA, Article 28).
- Identify applicable obligations under the European AI Act and NIS 2 if relevant, and propose associated measures.
- Contribute to client-facing documents: security questionnaire responses, security assurance plan (PAS), transparency note.
- Team Support
- Support developers and the Product Owner throughout the sprints: targeted architecture and code reviews, decision-making, DevSecOps best practices.
- Lead or supervise a penetration test / security audit before the pilot launch, and track fixes.
- Transfer knowledge so the team is autonomous after the mission.
- Resilience and Service Continuity
- Define, with the Product Owner and management, the RTO (maximum interruption duration) and RPO (maximum acceptable data loss) objectives. These are set per component: application, database, search index, LLM call, logging.
- Design a resilient architecture: encrypted, restorable, and tested backups, replication within the EU, management of AI provider dependencies (degraded mode, failover, or graceful refusal if the LLM is unavailable).
- Document recovery procedures (DRP/BCP): runbooks, roles, escalation chain, client communication.
- Conduct at least one restore test and one failover exercise before the pilot launch, then document the results and discrepancies.
- Verify that recovery mechanisms respect client segmentation and encryption (restore without tenant mixing, keys available in case of disaster).
Project details
- Start: flexible
- Remote: hybrid
- Location: Paris, France, FR
- Rate: 81-88 EUR/h
- Languages: French
Indexed by 9am's meta-search. You are a job provider? Claim or manage this listing
Found on collective.work