Lethal Autonomous Weapon

Definition

A lethal autonomous weapon (LAW), sometimes referred to as a lethal autonomous weapon system (LAWS) or “killer robot,” is a weapon system capable of independently selecting and engaging targets using sensor data and algorithms, without requiring real-time human authorization for each engagement. The degree of autonomy varies across a spectrum: from AI-assisted targeting systems where humans retain final strike authority (“human-in-the-loop”) to fully autonomous systems that can identify, track, and attack targets without human intervention (“human-out-of-the-loop”). Current deployed systems generally fall in the AI-assisted category, where AI processes intelligence data and recommends targets for human approval — though the quality and independence of that human review varies significantly in practice.

How It Relates to AI Threats

Lethal autonomous weapons represent a convergence of multiple AI threat domains. Within Systemic & Catastrophic Risks, LAWS raise questions about strategic stability, arms races, accountability for civilian casualties, and the irreversibility of lethal AI errors. Within Human–AI Control, the human oversight layer in AI-assisted targeting systems can be degraded through automation bias (operators rubber-stamping AI recommendations), institutional understaffing of review processes, or data quality failures that the AI cannot detect. The distinction between “AI-assisted” and “fully autonomous” targeting becomes less meaningful when the human oversight infrastructure is degraded to the point where AI recommendations are followed without genuine independent review.

Why It Matters

• AI-assisted targeting systems are already deployed by multiple militaries, making this an active rather than hypothetical threat category
• The accuracy of AI targeting is fundamentally constrained by input data quality — stale, incomplete, or misattributed intelligence data can lead to catastrophic civilian harm regardless of algorithmic sophistication
• Civilian protection depends on human oversight, but that oversight can be degraded through staffing cuts, time pressure, and automation bias — creating “autonomy by default” even in nominally human-in-the-loop systems
• International legal frameworks (International Humanitarian Law, Geneva Conventions) require human judgment in targeting decisions, but the practical meaning of “meaningful human control” remains contested

Real-World Context

Incidents in the TopAIThreats database involving AI-assisted military targeting include cases where stale intelligence data fed to targeting platforms contributed to strikes on civilian infrastructure (INC-26-0029), and where government pressure on AI companies to participate in military applications created tension with corporate safety commitments (INC-26-0028). International organizations including the UN, ICRC, Human Rights Watch, and Amnesty International have called for binding international protocols on autonomous weapons, with ongoing negotiations at the UN Convention on Certain Conventional Weapons.