Loading page…
The October 19 disclosure describes proof-of-concept work, not a documented attack. Its author argues that infrastructure and executive protection must account for autonomous targeting.
Listen to this story
Robert J. Bunker’s account of 2025 proof-of-concept work at C/O Futures points to a security risk beyond AI-generated information: coding assistance may help connect image recognition to a drone’s physical behavior. The potential targets he identifies include critical infrastructure, dignitaries, executives, and their vehicles. The account is qualitative, with no reliability figures or detailed methodology, and Bunker says no successful extremist, terrorist, or cartel attack of this kind has been publicly reported. He recommends combining counter-drone defenses with AI-focused security practices.
Bunker says the simulated group had no coding training and that he directed the work.
The setup linked generative-AI coding assistance with machine-vision models for facial and pattern recognition.
The account gives no success rate or detailed reliability testing beyond the described activities.
People without formal coding skills used generative AI to program consumer drones to identify and lock onto objects, researcher Robert J. Bunker says. In an October 19, 2026 article for The Soufan Center, he disclosed proof-of-concept work at C/O Futures and warned of new risks for critical infrastructure and prominent individuals.
Bunker is C/O Futures’ director of research and analysis and is affiliated with Arizona State University’s Future Security Initiative. He says the activities began in 2025 and simulated an extremist or terrorist group with no coding training. He directed the work himself.
The mechanism connects two kinds of AI. Generative AI supplies programming assistance; machine vision, meaning software that recognizes patterns in images, supports identifying objects. Bunker describes facial and pattern recognition using machine-learning models. In his account, the coding assistant helps a person without programming expertise connect those capabilities to a drone’s autonomous behavior.
His argument shifts attention from AI-generated information to software controlling a physical device. The article contrasts that risk with better-recognized malicious uses of generative AI, including propaganda, analysis and mission planning. Here, the concern is what generated code could enable a machine to do.
The article offers a qualitative description of the activities, not a published performance evaluation. It gives no success rate or detailed methodology for judging reliability outside those activities. The finding is an attributed demonstration of AI-assisted programming and recognition, not a measured estimate of an inexperienced attacker’s chances of success.
Bunker nevertheless argues that autonomous targeting could allow hostile groups to field what he calls “fire and forget” weapons: drones directed at a target without continued human piloting. His warning covers infrastructure, dignitaries, executives and the vehicles carrying them, rather than only military targets.
Bunker’s proposed response combines counter-drone measures with approaches aimed at AI-enabled threats. He calls for protection teams to draw on counterterrorism principles, adversarial testing and an understanding of machine vision and physical AI—software acting through machines. His recommendation brings those disciplines together rather than treating flying hardware and its guiding software as separate security problems.
Loading discussion...
Join the conversation
Explain what would justify acting before stronger evidence arrives.
Be the first to share a perspective or an experience.
Reader comments
Newest comments first. Replies stay oldest first.