The National Security Agency has released new guidance on managing risks across the AI supply chain, highlighting growing cybersecurity concerns tied to AI and machine learning systems. The joint information sheet outlines how organisations can better assess vulnerabilities when deploying or sourcing AI technologies.
The document defines the AI and machine learning supply chain as a combination of key components, including training data, models, software, infrastructure, hardware, and third-party services. Each element can introduce risks affecting confidentiality, integrity, or availability, particularly as advanced tools such as large language models and AI agents become more widely adopted.
Security risks associated with data include bias, poisoning attacks, and exposure via techniques such as model inversion and data extraction. For models, the guidance warns of hidden backdoors, malware, evasion attacks, and model manipulation. Organisations are advised to use trusted sources, perform integrity checks, and maintain verified model registries to mitigate such threats.
The paper also highlights software and infrastructure vulnerabilities, noting that AI systems often rely on complex dependencies that expand the attack surface. Recommended measures include malware scanning, testing, patching, and maintaining software bills of materials. Additional risks arise from third-party services, which may introduce weaknesses through their own supply chains or shared environments.
To manage these risks, organisations are urged to improve visibility across their AI ecosystems, identify suppliers and subcontractors, and require documentation such as AI and software bills of materials. The guidance aligns with frameworks from the National Institute of Standards and Technology and MITRE, reinforcing the need for coordinated approaches to AI supply chain security.
Would you like to learn more about AI, tech, and digital diplomacy? If so, ask our Diplo chatbot!
