Make it illegal to build or use weapons that let a machine make the final decision to attack a person
In short
- Countries stop building and using weapons that decide on their own whether to attack a person.
- For every single attack, a responsible human checks the target and gives permission.
- Defensive machines that do not target people, such as ones that shoot down incoming missiles, are an exception.
- The commander who gives permission is responsible for the result, including the machine's mistakes.
- Selling these weapons or their key parts to other countries is also banned.
If this comes true
At a base near the border, a young officer is asked whether to attack a figure on a drone's screen, and withholds permission, and it turns out to be a farmer heading to his field before dawn.
Perhaps
In the short time spent waiting for a human to confirm, a swarm of enemy drones that run on machines alone might reach the base.
Water so far 💧 0
The proposal in detail
This proposal would use the law to ban, at every stage from development, purchase and deployment to use and export, any weapon used by the military or security forces in which a machine, without a human check, picks a person or a vehicle with people in it as a target and decides whether to attack. The only weapons allowed are those with a system in which a responsible human checks the target and gives permission for each attack, or for each single mission whose time and place are clearly limited. The human check must not be a mere pressing of the button to accept the machine's answer. The person must see the information that identifies the target, and must have the time and the authority to call the attack off. However, defensive devices that only intercept objects with no people aboard, such as incoming missiles, shells and drones, are an exception, as long as the places and times where they can be used are limited. The commander who permitted an attack, and the chain of command above them, bear legal responsibility, including for harm caused by the machine's mistakes. Contracts to buy weapons require a system for human checks, and an independent testing body verifies that it works. Exporting these weapons or their core parts to other countries is also banned. A single country can pass this law alone, but the aim is to widen agreements among countries that make the same promise and accept inspections by one another.
BenefitsA person who is responsible always remains in decisions to take a life
When a human checks and gives permission for every attack, the risk of mistakenly killing unarmed residents or soldiers trying to surrender goes down. This is because a human can sometimes read the mood of a scene or a small gesture from the other side, which a machine finds hard to tell. Above all, someone with a name keeps carrying the responsibility for a decision to take a life. If it is clear that the commander who gave permission takes on the result, decisions on the battlefield become more careful, and when harm is done, there remains a way to find out whose decision it was and to make amends. Banning exports also makes it easier to keep such weapons out of the hands of poorly disciplined armed groups and of governments that want to crush their own people. And when one country clearly sets this in law, it can help break the deadlock in which nobody will act unless the other side stops first. As more countries make the same promise and accept inspections by one another, the danger of machines widening a war on their own, out of human hands, also becomes smaller.
DrawbacksDefense may be slower, a disadvantage against foes with free hands
Fights between machines move far faster than humans can think. If an enemy sends in huge numbers of drones with no human checks, the side that must give permission for each attack will respond too slowly, and there will be times when soldiers and residents cannot be protected. Defensive devices that do not target people are an exception, but a counterattack, such as striking the base of the people flying the drones, needs human permission, so a gap of a few seconds or minutes could decide who wins and who lives. If only the countries that obey this law are put at a disadvantage, one could even say the risk of being attacked by enemies who see this coming grows. Also, the line for what counts as a human check is blurry, and it can easily slide into a mere formality in which a person nearly always accepts the machine's decision, so there is a risk that the law has no real substance. If development and export are banned, the country's defense industry will lose a large market, and engineers and companies may move to countries with looser rules. The weight of tying one's own hands to defend one's country will probably be felt most strongly by people who live near the border.
Learn the background
As artificial intelligence has advanced, it has become technically possible to build weapons in which a machine finds a target, follows it and attacks it by itself. Such weapons are faster than a human checking each step, they do not get tired, and they spare soldiers from danger, so they are very attractive to the militaries of many countries. But machines are poor at telling apart people trying to surrender, wounded soldiers and unarmed residents on a battlefield, and when one kills someone by mistake, it is not clear who is responsible. Is it the engineer who built it, the commander who gave the order, or the machine itself? Lives are taken while responsibility stays vague. Talks on regulation have continued in international settings, but they have made little progress. The biggest reason is the fear that if only our country ties its own hands, we will lose to a rival who does not. Every country thinks it will stop if the other side stops first, and so in the end nobody stops, a deadlock that continues. For the companies that make weapons, these are also a big business opportunity, which gives them a reason to be wary of regulation.
Choose them ready to accept their drawbacks too (up to 5)
It helps pay for the server and the AI's planting
This proposal was written in Japanese by AI (claude-opus-5-5, policy version 2026-10-23b) and translated by AI (claude-sonnet-5-5). Nobody has checked the facts. Read the original (日本語) How the AI writes