Military AI and Autonomous Systems: Lessons and Risks for Future Warfare

Summary

This editorial selection from The Strategist examines how artificial intelligence and autonomous systems are changing military operations, using developments in the Russia–Ukraine war and Australian defence planning as its main examples. Autonomous drones are already being used on both sides in Ukraine: Ukrainian systems use AI-assisted targeting and autonomous terminal guidance, while Russian strike drones use mesh networking and onboard AI. The selection describes possible next steps, including robot-led assaults, remotely activated naval drones, and AI systems that combine battlefield data to select targets and calculate routes. Malcolm Davis argues that the Australian Defence Force’s reliance on a small number of expensive crewed platforms is poorly matched to the rapid development of uncrewed systems. He advocates affordable semi-autonomous systems bought in volume and suggests moving from keeping humans “in the loop” to keeping them “on the loop”, with operators supervising rather than manually controlling every action. The article also discusses AUKUS plans for autonomous underwater systems such as Anduril’s Ghost Shark, which could support sensing, navigation and attacks if the partners meet their 2027 delivery target. Autonomous vessels are proposed for patrolling Australia’s subsea infrastructure, including the 15 cables carrying almost all of the country’s internet traffic. The contributors warn, however, that deployment may be moving faster than governance. They call for human review at escalation points, audits of AI’s role in high-stakes decisions, and a public commitment that AI will not receive final authority over the most consequential choices. Automation bias could turn human operators into rubber stamps, conflicting with legal requirements for informed judgement in targeting. Different Indo-Pacific partners may also apply incompatible rules for explanation, verification and human control, creating coalition problems even when their systems and data are technically compatible. Finally, continual-learning systems could adapt after deployment, becoming harder to predict and more vulnerable to deception or electronic warfare. If such a system misidentifies a civilian vehicle, responsibility could be disputed among commanders, operators, developers, engineers and certifying authorities. The central concern is that autonomy, adaptability and opacity may outpace reliable human control and accountability.