AI glossary · Basic · Safety & alignment · Robotics & physical AI · Defense & security
Autonomous weapon systems
Lethal autonomous weapon systems (LAWS) are weapon systems that, once activated, can select targets and carry out attacks without a human approving each action. Whether and how to regulate them remains an ongoing debate in the international community.
About this term
- Level
- Basic · Technical terms you often meet in AI news
- Category
- Industry AI
- Industries
- Defense & security
- Added
- 2026-10-10
What it is
The key question that defines an autonomous weapon system is 'who makes the final decision to attack.' Drones remotely piloted by humans are not considered autonomous weapons. An approach where a human approves every action is called human-in-the-loop; one where the machine decides and a human watches and can only stop it is called human-on-the-loop; and one where humans are removed entirely is called human-out-of-the-loop. The debate usually centers on the last stage, where targets are selected and attacked without human intervention.
How much are humans involved?
- Remote control
- Human-in-the-loop
- Human-on-the-loop
- Human-out-of-the-loop
Why it matters
As AI has become better at image recognition and decision-making, the range of autonomy that is technically possible has widened. Supporters say such systems react faster and can reduce risk to soldiers. Opponents argue that machines deciding over life and death is itself contrary to human dignity, and that it is difficult to uphold the principle of international humanitarian law that distinguishes civilians from combatants. For this reason, definitions and regulatory approaches have been discussed within international frameworks such as the UN Convention on Certain Conventional Weapons (CCW).
Things to know
There is still no single internationally agreed definition, so countries may disagree on whether the same system counts as an autonomous weapon. A frequently raised issue is the 'accountability gap': when a malfunction or wrong decision occurs, it becomes unclear whether the developer, the commander or the operator is responsible. AI models can behave unpredictably in situations they were not trained on, so the risk grows in environments with many variables, such as a battlefield. As a result, much of the discussion converges on how to guarantee 'meaningful human control.'
Example
In defense, AI assistance tools are used to find and flag suspicious objects in surveillance footage, while humans decide whether to actually respond. The same recognition technology is a decision-support tool if it is designed so that humans make the decision, but it becomes a subject of the autonomous weapons debate if it is designed to go all the way to attack without a human. That is why countries' defense AI principles documents are often written around 'where and how humans make judgments' rather than the technology itself.