- AI systems may create national security risks even outside national security domains through misuse, scalability, generative capabilities, corrupted data, or software vulnerabilities.
- The addition prioritizes the NIST AI RMF Playbook’s Map function because contextual knowledge is needed to identify risks and broader contributing factors before risk management can be effective.
- The document cites sensitive data exposure risks from geospatial mapping, public fitness tracking, and large-scale cell-phone location analytics that can identify sensitive facilities and individuals’ homes.
- Suggested actions call for interdisciplinary teams that include national security, human factors, socio-technical, legal, public policy, systems design, and engineering expertise.
- Organizations are advised to consider off-label uses, adversary misuse, deployment in different geopolitical environments, and how AI capabilities could be composed into other applications with national security implications.
This April 2023 SCSP document supplements the National Institute of Standards and Technology Artificial Intelligence Risk Management Framework Playbook with national security considerations. It frames the NIST AI RMF as a voluntary tool for managing AI-associated risks to individuals, organizations, and society, and argues that AI systems can pose national security threats even when not designed for national security use cases. The resource is organized around the RMF Playbook’s Map function, with added considerations, suggested actions, transparency and documentation questions, and references for selected Map subcategories. It highlights risks including exposure of sensitive locations through geospatial, fitness-tracking, and cell-phone data; deepfake impersonation; extreme scalability; generative capabilities; corrupted data or software; off-label use; adversarial exploitation; and third-party supply chain dependencies. The document emphasizes early system-requirements work, interdisciplinary expertise, stakeholder engagement, risk-tolerance documentation, testing of human-AI configurations, mapping of external connections, and review of foreign-controlled technologies or components. Its policy relevance lies in helping AI developers, deployers, and users incorporate national security context into risk identification and assessment under the NIST AI RMF, especially when documenting deployment contexts, impacted stakeholders, third-party resources, and emerging risks across the AI lifecycle.
National security threats may be posed by AI systems even if they were not created for the national security domain.
SCSP
- Inventory third-party material required for system implementation and maintenance. Determine national security risks from critical third-party resources that are impacted by foreign entities as a supply chain vulnerability.
- Establish risk criteria in consideration of different sources of risk and different levels of risk. Review uses of AI systems for “off-label” purposes, especially in settings that organizations have deemed as high-risk.
- Plan and test human-AI configurations under close to real-world conditions and document results. Include examples of “off-label” use, “on-label” use with unintended consequence, and potential “misuse cases” by bad actors.
- Consider narrowing contexts for system deployment. Consider how outcomes may directly or indirectly affect users, groups, communities, national security, and the environment.
Published by the Special Competitive Studies Project. Co-publishers and external authors, where they exist, are credited in the byline above.