China develops autonomous lethal drone swarm algorithm
Chinese researchers developed an AI algorithm enabling autonomous drone swarms to independently identify and kill targets in contested environments without human control or communication, reportedly achieving 100 per cent target elimination in simulations. The technology points toward future military deployment of fully autonomous lethal drone swarms with minimal human oversight.
What happened
Chinese researchers at Northwestern Polytechnical University developed an AI algorithm called HG-STR (Heterogeneous Graph Spatio-Temporal Reasoning) designed to enable autonomous drone swarms to locate and eliminate targets without human control.
The system allows drones to operate independently in contested environments where communications are jammed, making targeting decisions in 6.6 milliseconds based on incomplete information from individual drones in the swarm.
According to the peer-reviewed paper published in Acta Aeronautica et Astronautica Sinica on May 19, 2026, the algorithm achieved a 100 per cent target elimination rate in simulation tests of a 10-drone swarm across a 100 km × 100 km area.
Reported impact
A Beijing-based defence expert stated the technology enables drone swarms to be deployed with only a final order to 'find and kill' designated targets, operating independently in high-risk combat zones after communications are cut off.
The system's speed advantage—calculating decisions in 6.6 milliseconds compared to several seconds for older methods—is presented as eliminating decision delays that could otherwise leave drones vulnerable.
Index analysis
The algorithm uses a heterogeneous graph system that assigns different categories and priorities to battlefield elements, distinguishing between friendly forces, enemy targets and terrain, rather than treating all data uniformly.
The system incorporates a compressed memory function allowing drones to continue reasoning from past observations even when communication systems fail, reducing reliance on real-time data exchange.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.