Autonomous Weapons Systems

Autonomous weapons systems are weapon systems that can search for, identify, track and engage targets without meaningful human control over individual attacks. Unlike a homing missile, which is locked onto a target by a person, an autonomous system chooses its own targets based on programmed criteria and sensor data. The prospect of delegating life-and-death decisions to software has made the topic one of the most contested areas of AI policy.

History and Background

Loitering munitions and automated air-defence systems have existed for decades, but public debate began in earnest around 2012–2013, when advances in computer vision and machine learning made target recognition reliable enough to contemplate full autonomy. In 2013 the Campaign to Stop Killer Robots launched its call for a new treaty requiring meaningful human control over the use of force.^[1]

Discussions moved to the UN Convention on Certain Conventional Weapons (CCW), where a Group of Governmental Experts on lethal autonomous weapons systems (GGE on LAWS) has met almost annually since 2016, producing a "rolling text" of possible provisions. In 2025 the pace increased: a cross-regional group of 42 states declared itself ready to begin treaty negotiations, and in November 2025 the UN General Assembly's First Committee adopted a resolution on autonomous weapons for the third consecutive year, supported by 156 states.^[1][2] The resolution was adopted by the General Assembly as Resolution 80/57 in December 2025, explicitly referencing the Secretary-General's call for a legally binding instrument built on prohibitions plus regulations.^[2]

Key Concepts

• Definition problem. There is still no agreed UN definition. States differ on whether to ban systems that cannot be used with human control, or all systems whose attacks humans do not individually approve — a distinction that determines whether existing air-defence systems are covered.^[3]
• Meaningful human control. The concept at the centre of the debate: a human must understand enough about a system's functioning and context to take moral and legal responsibility for an attack. Critics argue that fast-moving, machine-speed engagements make genuine human judgement impossible.^[1]
• Existing autonomy. Some degree of autonomy is already deployed — Israel's Iron Dome, Russia's Pantsir, naval close-in weapon systems and loitering munitions — which is why advocates frame the issue as regulating a trend, not banning a hypothetical.^[3]
• Legal accountability gap. International humanitarian law requires distinction, proportionality and precaution. If an algorithm errs, it is unclear who bears responsibility: the commander, the operator, the manufacturer or the state.^[4]
• Proliferation and arms races. UN resolutions cite risks of an emerging arms race, lowered thresholds for conflict, miscalculation and transfer to non-state actors.^[2]

Applications and Impact

The technology is not hypothetical: autonomous target selection already exists in certain defence systems, and commercial drones with AI-based person detection are widely available. The policy trajectory through 2026 centres on the CCW's Seventh Review Conference in Geneva (late August–early September 2026) and the closing sessions of the GGE mandate, which many states want converted into a negotiating mandate for a binding treaty.^[1][3] Meanwhile, major military powers are investing in autonomous and semi-autonomous systems regardless of the outcome, so the practical question for the rest of the decade is less whether such systems will exist than which uses are declared unlawful.

>See Also

• AI Ethics
• UN AI Governance
• OECD AI Principles
• AI Safety
• Malaysia AI Governance Framework
• EU AI Act