Safety and War: Safety and Security Assurance of Military AI Systems
Emerging scholarly pressure on military AI safety assurance could shape international standards and export control frameworks that Australian Defence and DISR must navigate.
Key points
- AI Now Institute argues military AI systems like Lavender and Gospel lack adequate safety assurance or independent oversight.
- The paper calls for safety engineering frameworks to evaluate military AI fitness-for-use against defined risk thresholds.
- Limited direct APS applicability now, but relevant to Defence AI governance and export control policy discussions.
Summary
A new AI Now Institute publication argues that military AI systems are being developed and deployed without adequate safety engineering assurance, independent evaluation, or compliance with international human rights law. Using Israeli AI targeting systems Lavender and Gospel as case studies, the authors identify a critical gap between claimed capabilities and verified performance. The paper advocates applying safety assurance methodologies - claim-oriented, use-case-specific evaluation - to military AI, and flags export controls and global governance bodies as key policy arenas. Future posts will address conceptual safety grounding and specific policy mechanisms.
Implications for Australian agencies
- Monitor Defence, DISR, and DFAT policy teams may want to monitor AI Now's forthcoming work on military AI export controls, as it could inform international negotiations Australia participates in.
- Consider Agencies working on AI risk frameworks could consider whether safety assurance gaps identified in military contexts have analogues in high-stakes civilian government AI deployments.
Implications are AI-generated. Starting points, not advice.
"Safety and War: Safety and Security Assurance of Military AI Systems" Source: AI Now Institute – Publications Published: 25 June 2024 URL: https://ainowinstitute.org/publications/safety-and-war-safety-and-security-assurance-of-military-ai-systems A new AI Now Institute publication argues that military AI systems are being developed and deployed without adequate safety engineering assurance, independent evaluation, or compliance with international human rights law. Using Israeli AI targeting systems Lavender and Gospel as case studies, the authors identify a critical gap between claimed capabilities and verified performance. The paper advocates applying safety assurance methodologies - claim-oriented, use-case-specific evaluation - to military AI, and flags export controls and global governance bodies as key policy arenas. Future posts will address conceptual safety grounding and specific policy mechanisms. Implications for Australian agencies: - [Monitor] Defence, DISR, and DFAT policy teams may want to monitor AI Now's forthcoming work on military AI export controls, as it could inform international negotiations Australia participates in. - [Consider] Agencies working on AI risk frameworks could consider whether safety assurance gaps identified in military contexts have analogues in high-stakes civilian government AI deployments. Retrieved from SIMS, 18 May 2026.