In Ukraine, cheap drones are destroying weapons worth many times more. Japan has decided it cannot afford to keep watching from the sidelines.
Japan’s Cabinet on Tuesday adopted the country’s 2026 defense white paper, calling for faster adoption of combat drones and AI-assisted military systems.
Turning policy into combat power will test how far Tokyo is willing to trust the machines.
A drone force designed to watch and strike
Under the SHIELD plan, defense officials want to deploy large numbers of low-cost aerial drones, surface vessels, and underwater vehicles around the coast.
Several systems have direct combat roles. Portable drones would detect and strike invading vehicles, while ship-launched models could attack hostile warships or intercept drones approaching radar sites.
Defense officials have chosen four interceptor-drone systems for testing, including Terra Drone’s TerraB1. An Associated Press report puts the original field at 38 applicants. Defense Minister Shinjiro Koizumi summed up the timetable as “Speed is key.”
AP also reported that Toshiba plans to develop combat drones with Japanese startup Prodrone.
Ukraine’s drone war reaches the Pacific
Ukraine is teaching military planners how quickly software can change combat. According to defense officials, AI and military networks can speed decisions, while front-line data allows developers to update software as fighting continues. Ukraine’s drones and ground robots already follow this rapid test-and-update cycle.
Closer to home, Tokyo cites increased Chinese naval operations and military exercises around Taiwan. Beijing’s development of AI-enabled combat robots gives the country another reason to accelerate its unmanned programs.
A shrinking population compounds the pressure. A defense transformation plan says unmanned systems have become increasingly urgent. Drones could keep troops out of dangerous missions while allowing fewer personnel to monitor larger coastal areas.
Who controls the attack?
Japan has ruled out lethal autonomous weapons operating entirely without human involvement. Its responsible AI guidelines for defense systems require human judgment, but SHIELD plans do not say when an operator must approve a strike or how control would hold up if communications were jammed.
Before deployment, vendors must turn that policy into controls an operator can actually use. Records should identify the data behind each AI recommendation and show when a person intervened. Ukraine’s Palantir-backed military AI lab offers an early comparison, using restricted access to battlefield data and repeated system testing.
An audit trail can help reconstruct a decision later. During an attack, commanders could still find themselves rushing through possible targets without seeing how uncertain the software is.
Failures would reach far outside the command room. Residents near future test or deployment sites would live with the consequences of lost communications or a misidentified target. Until officials explain who can authorize a strike and who answers when AI gets it wrong, Tokyo’s promise of human control remains incomplete.
Chinese military-linked institutions reportedly used outputs from OpenAI and Anthropic models to train domestic defense systems.