Adversarial-AI vulnerability assessment for defence using AI security analysis
Hardening military AI against adversarial attack.
The Challenge
As NATO and its member nations adopt AI across military systems — from situational awareness to decision support — those systems inherit a new attack surface: adversarial-AI techniques such as data poisoning, model evasion, spoofing and perturbation attacks can silently corrupt the outputs commanders rely on. NATO needed a clear-eyed assessment of these vulnerabilities and practical mitigations before AI is trusted in operational defence settings.
The Solution
NQM was engaged in a study for NATO (delivered via Frazer-Nash Consultancy, alongside Copper Horse) to identify and mitigate the security vulnerabilities of AI when applied in military and defence settings. NQM brought its applied adversarial-AI resilience experience — including the trust-scored, cyber-resilient data-fusion approach developed for maritime combat-system analytics on MAXAD — to the assessment of adversarial-AI threats and the framing of mitigations relevant to defence AI systems.
Outcomes
The engagement produced analysis of adversarial-AI threats and proposed mitigations for defence AI deployments. INPUT-NEEDED: formal deliverables/outcomes to be confirmed from internal records. NOTE: this engagement is distinct from NquiringMinds’ separate participation in the NATO Innovation Continuum (IGNITE/SHINE) activities.
Features
Identification of security vulnerabilities of AI applied in military and defence settings.
Assessment of adversarial-AI threats (e.g. poisoning, evasion, spoofing) relevant to defence systems.
Proposed mitigations applicable to NATO and member-nation defence AI deployments.
Assessment grounded in NQM's operational experience hardening deployed defence AI (MAXAD/DFCR line of work), not desk research alone.
Benefits
Provided NATO (via Frazer-Nash) with an assessment of adversarial-AI vulnerabilities and mitigations for defence AI systems (delivered engagement).
Strengthened NQM's defence AI-security credentials alongside MAXAD and BAE INTeACT IAMD.
