17.15 National security
National security AI applications span surveillance, autonomous weapons, cyber-defence and intelligence analysis. The space is opaque by nature; this section describes what is publicly known.
Autonomous weapons. The debate over lethal autonomous weapon systems (LAWS) has been ongoing at the UN Convention on Certain Conventional Weapons (CCW) since 2014, without producing a binding international treaty. The Stop Killer Robots coalition has campaigned for a ban; the major military powers (US, UK, China, Russia) have not supported one. Practical deployment has continued: loitering munitions with target-recognition capability (Israel's Harop, used in the 2020 Nagorno-Karabakh conflict and by Ukraine), drone swarms, AI-enabled missile-defence systems. Project Maven, the US Department of Defense initiative to apply ML to drone-footage analysis, was begun in 2017, prompted Google employee protests in 2018 leading to Google's withdrawal, and continues with other contractors. The US Replicator initiative, announced by Deputy SecDef Kathleen Hicks in August 2023, is intended to field thousands of attritable autonomous systems by mid-2025.
Surveillance. Facial-recognition deployment varies enormously by jurisdiction. Clearview AI's database (over 50 billion images scraped from the public internet) and its provision to US law enforcement attracted regulatory action in multiple jurisdictions, including UK ICO fines and EU GDPR enforcement. China's deployment of facial recognition in public space, particularly in Xinjiang, has been documented in detail and has been the subject of US sanctions on companies including SenseTime and Megvii. The EU AI Act largely prohibits real-time public-space facial recognition with narrow exceptions; the US has no comparable federal restriction.
Cyber-defence. ML-based intrusion detection, anomaly detection in network traffic, and malware classification have been used at scale for over a decade. The integration of LLMs has accelerated both offensive capability (automated reconnaissance, social-engineering at scale, vulnerability discovery) and defensive capability (faster triage, better detection rules, automated patching). Microsoft's 2024 announcement that it was using AI to detect zero-day vulnerabilities in widely used libraries, and the various Project Zero–style efforts at major tech firms, illustrate the dynamic.
Intelligence analysis. Palantir, Anduril, Scale AI, and the various other defence-AI vendors have grown substantially in revenue and prominence in 2023–2025 alongside increased US and allied defence spending and the wars in Ukraine and the Middle East. The US Department of Defense's 2023 deal with Scale AI for the Thunderforge programme, Palantir's role in the IDF and the Pentagon, and Anduril's drone and counter-drone systems are publicly documented examples.
The geopolitical dimension dominates discussion of the next decade. The October 2022 and October 2023 US export controls on advanced AI chips to China, the corresponding Chinese investment in domestic AI capability (Huawei Ascend, the 2024–2025 advances of DeepSeek and other Chinese labs), and the broader US-China technology competition shape every aspect of the field.