Building trust in AI through a cyber risk-based approach
This joint high-level risk analysis provides a cyber risk-based approach to artificial intelligence, published by ANSSI and co-signed by international cybersecurity authorities ahead of the Paris AI Action Summit. It examines AI-specific vulnerabilities such as data poisoning, model extraction, and evasion attacks, alongside general IT infrastructure risks and supply chain challenges. Finally, it outlines practical recommendations for AI users, operators, developers, and policy-makers to secure AI systems and foster trust.
Summary and content fields are drawn from the document text and are pending editorial review.
Who issued it
Who it governs
This document is for AI developers, operators, users, and policy-makers seeking to understand and mitigate cyber risks in AI systems.
What the document contains
The document addresses cybersecurity risks associated with artificial intelligence, outlining AI-specific attack vectors such as data poisoning, model extraction, and evasion. It covers supply chain security, infrastructure vulnerabilities, lateralization risks, and human/organizational failures like shadow AI. Finally, it offers practical implementation guidelines and self-assessment checklists for AI developers, users, operators, and policy-makers.
The document does not provide legally binding regulations or enforcement mechanisms, nor does it detail AI-enhanced cybersecurity defensive or offensive software solutions, which are noted as being outside its scope.
- AI systems are software systems that face general information system vulnerabilities alongside AI-specific threats.
- AI-specific cyber attacks fall mainly into three categories: poisoning, extraction, and evasion.
- Securing AI requires mapping and protecting the entire AI supply chain, including compute, models, and data.
- The opacity of many AI systems ('black boxes') complicates root-cause error analysis and incident investigation.
- Policy-makers and organizations must collaborate internationally to establish shared standards, evaluation capacities, and continuous threat monitoring.
· Building trust in AI through a cyber risk-based approach · Challenges and scope of an AI risk analysis · Key risks and attack scenarios · Guidelines for AI users, operators and developers · Guidelines for policy-makers · Appendix 1: Recommendations for the secure implementation of an AI system · Appendix 2: References
How it scores
Each score is backed by a quote verified against the document text. Editorial review has not yet confirmed them.
Accountability
2/5Clear responsibility for AI outcomes
F oster dialogue between cyber and AI actors, in particular between cybersecurity agencies and AI Safety Institutes (or similar), while clearly defining respective perimeters and responsibilities
GUIDELINES FOR POLICY-MAKERS
screening score · 2026-10-23
Human Dignity
0/5Respect for the inherent worth of persons
screening score · 2026-10-23
Privacy
4/5Protection of personal data and private life
design the AI system using a privacy by design approach to meet data protection requirements throughout the lifecycle: take into account data confidentiality issues
APPENDIX 1
screening score · 2026-10-23
Enforcement
0/5Consequences and compliance mechanisms with real force
screening score · 2026-10-23
Impact Assessment
4/5Required assessment of an AI system's risks/impacts before or during use
Such a deployment should therefore include a dedicated risk analysis to assess the risks and identify appropriate security measures.
CHALLENGES AND SCOPE OF AN AI RISK ANALYSIS
screening score · 2026-10-23