First published June 14, 2021.
In March 2021 a United Nations panel of experts on Libya reported that a Turkish-made Kargu-2 drone had "hunted down and remotely engaged" retreating fighters during the country's civil war in 2020, and described the drone as a lethal autonomous weapon. When the report drew public attention in May and June 2021, headlines called it the first time a robot had killed people on its own. The report itself did not go that far: it did not confirm that anyone died in an autonomous attack, and key details of how the drone was operated remain unknown.
Even so, the episode became a landmark in the debate over weapons that can choose and attack targets without a human decision. Here is what the report said, what it left open, and why the argument over "killer robots" still matters.
What the UN panel reported
The UN Security Council's Panel of Experts on Libya monitors the arms embargo on the country. Its report, dated 8 March 2021, covered fighting in 2020 between forces of the Tripoli-based Government of National Accord (GNA) and the Libyan National Army (LNA) led by Khalifa Haftar.
The panel wrote that as Haftar-affiliated units retreated, they were pursued by drones and loitering munitions, including the Kargu-2. It described the system as a lethal autonomous weapon and, according to analysts who studied the report, said it could be programmed to attack without a live data link to its operator. Remains of a Kargu-2 were later recovered in Libya.
What the report did not say is just as important. It implied that the drones may have caused casualties but did not state that anyone was killed by a Kargu-2 acting autonomously. It also did not establish whether the drone was used in a fully autonomous mode in that specific incident or under human control.
Why the wording matters
Zachary Kallenborn, a researcher who studies drone swarms and emerging weapons, was one of the first analysts to highlight the report. His assessment was carefully conditional. Speaking to NPR, he said: "If anyone was killed in an autonomous attack, it would likely represent an historic first known case."
The "if" is the heart of the matter. Three separate questions are often blurred together:
- Capability: can the Kargu-2 find and strike targets without a human decision?
- Use: was it actually operating that way in Libya in 2020?
- Outcome: did anyone die as a result?
The report offered evidence on the first, suggested the second and left the third unconfirmed. That is why "first robot killing" headlines overstated the case, even though the report was genuinely significant.
What the Kargu-2 is
The Kargu-2 is made by STM, a Turkish defence company. It is a small, portable quadcopter about 60 cm across, weighing around 7 kg, and designed to loiter over an area before diving onto a target with an explosive charge. STM lists a range of about 10 km and a flight time of around 30 minutes.
The company says the drone uses machine learning to identify and track targets, and STM's chief executive has claimed it can use facial recognition. However, STM's promotional videos show a human operator selecting targets before the drone attacks, and the company has never publicly demonstrated a swarm autonomously choosing and coordinating strikes. Claims about fully autonomous attacks rest mainly on the UN report and later media coverage.
Turkish forces have reportedly used the drone in Syria and Libya. Rumours that Azerbaijan used Turkish AI-enabled drones against Armenian forces in the 2020 war over Nagorno-Karabakh (Artsakh) circulated widely, but they have not been confirmed.
Who was fighting whom in Libya
Libya's civil war drew in many outside powers. Turkey, alongside Qatar, backed the UN-recognised GNA in Tripoli and supplied drones, equipment and military advisers. Haftar's LNA was supported by the United Arab Emirates, Russia (including Wagner Group mercenaries) and Egypt, among others. UN experts repeatedly found that states on both sides were violating the UN arms embargo.
The report heightened concern among Turkey's allies as well. Ziya Meral of the Royal United Services Institute, a British defence think tank, told an audience in London that it was time to assess where Turkish drones and advanced warfare technology stood and what they meant for the region and for NATO.
The case against, and the case for
Critics of autonomous weapons raise several concerns:
- Distinguishing civilians from fighters. In 2015 more than 1,000 AI researchers and public figures, including Stephen Hawking and Elon Musk, signed an open letter calling for a ban on offensive autonomous weapons and warning of an AI arms race.
- Flawed training data. Kallenborn has written in the Bulletin of the Atomic Scientists that machine-learning systems can "learn" the wrong lesson from datasets that are not complex or robust enough, and that their decisions are often opaque.
- Deception. Adversaries can deliberately fool image-recognition systems, for example with altered markings.
- Accountability. When a machine selects a target, it becomes harder to say who is responsible for a mistake.
Supporters argue that autonomy may be essential for countering very fast threats such as drone swarms, and that well-designed systems could make fewer errors than tired or frightened human operators, potentially reducing civilian harm. Kallenborn himself suggests judging risk by the chance of an error multiplied by its consequences, which depends on factors such as the targets chosen, the payload, the environment and how thoroughly a system has been tested.
Where the debate stands
Governments have discussed lethal autonomous weapons under the UN Convention on Certain Conventional Weapons since 2014, through a formal group of governmental experts set up in 2016. Topics include whether certain systems should be banned and how much human control international humanitarian law requires.
In December 2023 the UN General Assembly adopted its first resolution on the subject, with 152 countries in favour, 4 against and 11 abstaining, supporting continued international discussion. Campaign groups such as Stop Killer Robots, formed in 2013, are pushing for a binding treaty, while several major military powers prefer non-binding guidelines.
The Libyan case will likely remain a reference point in those talks, not because it proved that a robot killed someone, but because it showed how close the technology has come and how hard it can be to know what really happened on the battlefield.
Sources: Zachary Kallenborn, Bulletin of the Atomic Scientists (May 2021); VOA: Possible first use of AI-armed drones triggers alarm bells; Wikipedia: STM Kargu; Wikipedia: Lethal autonomous weapon; original reporting by NPR (June 2021).
COMMENTS