Launched by Campus Cyber, this working group – jointly led by HeadMind Partners and Hub France IA – comprises two sub-groups:
1. “Developing secure AI/Security by Design for AI systems”, co-led by HeadMind Partners and BNP Paribas.Its aim is to review the first version of the ANSSI guide published in April 2024 on the security of generative AI-based systems, to identify areas of confusion for experts unfamiliar with AI, the challenges to be anticipated during implementation, and any requirements that may need to be added.
2. “Analysis of Attacks and Defences on AI Systems”, co-led by the Gendarmerie Nationale and Hub France IA, this group aims to develop an operational approach to securing AI systems, drawing in particular on the work of ANSSI, NIST, MITRE (ATT&CK and ATLAS) and OWASP.This approach, structured around the cycle of Understand → Identify → Qualify → Protect → Remediate → Manage Risk, has resulted in a white paper on attacks and defences for AI systems.Version 3 is currently online and version 4, due to be released this autumn, will include:
- An educational guide to understanding the main attacks against AI systems, their evolution, their specific characteristics and how they fit into the lifecycle of AI systems;
- An operational reference framework in the form of attack fact sheets, enabling the identification, for each threat, of the assets and life-cycle phases affected, the impacts, the technical feasibility, and the associated prevention, detection and remediation measures;
- A business risk analysis method enabling organisations to move from identifying an attack to assessing its consequences for the organisation, taking into account applicable requirements and determining the level of residual risk.
