Required skills
- AI Security
- Generative AI Security
- Agentic AI Security
- AI Risk Analysis
- Offensive AI Testing
- LLM Red Teaming
- Prompt Injection
- Jailbreaking
- Data Poisoning
- AI Architecture Security
- AI Safeguards
- AI Governance
- Threat Modeling
- MITRE ATLAS
- OWASP Top 10 for LLM Applications
- Python
- MLSecOps
Languages: French
Job description
Our client, a large insurance group, is deploying generative and agentic AI uses and needs to master the associated risks. They are looking for an expert to secure these systems and assess their robustness against attacks.
Project overview
Context
Our client, a large insurance group, is deploying generative AI and agentic AI uses across its operations and needs to master the end-to-end risks. Its group cybersecurity department is looking for an expert capable of securing these systems, but also of assessing their robustness against attacks.
The consultant analyzes the risks of AI projects, tests models and agents offensively, and defines safeguards (segmentation, access control, deterministic controls) in conjunction with data, business, and IT teams. The team is looking for exceptional profiles: the level of experience matters less than proof of competence (CTF, research, publications, personal projects, competitions).
Missions
Tools & Environment
Business and Regulatory
Technical
Methodology and Tools
Documentary
Points to Clarify
- End-to-end risk analysis of generative and agentic AI projects
- Offensive testing of LLMs (prompt injection, jailbreak, data exfiltration, poisoning)
- Securing agentic architectures (limited autonomy, segmentation, deterministic authorizations)
- Defining safeguards and security standards for AI
- Governance of agent identities, rights, and actions, associated traceability
- Assessing AI risk from suppliers and third parties
- Monitoring threats and attacks related to AI
- Raising awareness among teams and contributing to the AI security policy
- Security of generative and agentic AI
- AI risk analysis
- AI Act, NIST AI RMF, ISO 42001
- Third-party risk management for AI suppliers
- Red teaming of language models
- Prompt injection, jailbreak, data poisoning
- Security of RAG architectures
- Security of agents and associated tools (function calling, MCP)
- Input and output safeguards and filtering
- Non-human identity management
- Security of cloud AI platforms (Azure AI, AWS Bedrock, Vertex AI)
- MLSecOps
- Python
- OWASP Top 10 for LLM Applications
- MITRE ATLAS
- Garak, PyRIT or equivalent
- Threat modeling applied to AI
- AI risk analysis reports
- AI security standards and guides
- The report lists "penetration testing" and "pentester" as two distinct needs: AO 1 application (web, API, mobile) and AO 2 infra (AD, network, cloud). If Xavier considers them as a single profile, they will need to be merged.
- The identified LinkedIn profiles are mainly oriented towards GRC and vulnerability management (DORA, ISO 27001, risk analysis, phishing). Only one has an offensive certification (HTB CWES). None were selected for the offensive AOs: the "gems" criteria come from the report (CTF, bug bounty, certifications, personal projects). The AI angle comes from his reposts (security of agentic systems, AI risk from suppliers).
Project details
- Start: flexible
- Remote: hybrid
- Location: Paris, France, FR
- Languages: French
Indexed by 9am's meta-search. You are a job provider? Claim or manage this listing
Found on collective.work