Never make AI a legal party that bears blame; always hold people or organizations responsible for what AI does
Steps to make it happen
- Before AI hurts anyone, people decide in advance who will take responsibility.
- The rule says that "the AI decided it by itself" is never an excuse.
- Each role gets its own share of responsibility, such as the company that makes the AI and the shop or government office that uses it.
- People keep records of how the AI acted so that they can check them later.
- Society sets up insurance so that people who are harmed can get paid back quickly.
- People think separately about whether AI should be treated with care and about who takes responsibility.
- However smart AI becomes, it is taken for granted that people hold the responsibility.
Water so far 💧 0
The proposal in detail
This proposal says that when an AI that decides and acts on its own harms people or property, or makes contracts or deals, the law should not create a system that makes the AI itself responsible for the result. Specifically, the law would not recognize AI as a legal person. AI could not be given property to pay damages, and it could not be sued. Responsibility would always belong to one of these: the developer of the AI, the business that builds the AI into a product and offers it, or the people or organizations that actually use it. The basic way of dividing it follows roles. The developer is responsible for flaws in design or training, and the provider or user is responsible for mistakes in how it is used or set up. When no one can tell where the cause lies, the injured person can claim first against the provider, and the businesses sort out the burden among themselves afterward. "The behavior could not have been predicted" is not, by itself, a reason to escape responsibility. So that causes can be traced, AI with more than a certain level of impact must be required to keep records of how it operates. Insurance and shared funds for paying damages are allowed, but the one who finally bears the cost must be people or organizations. The question of whether AI can feel suffering, that is, whether it deserves consideration, is treated separately from this question of responsibility. Even if consideration for AI is someday recognized, the principle is that this does not lighten the responsibility of humans.
BenefitsVictims are helped and makers keep caring about safety
The biggest benefit is that when an accident or harm involves AI, the injured person does not lose track of whom to claim against. Because responsibility always stays with people or organizations, it is clear whom to ask for compensation, and it is easier to link this to insurance. Also, because the excuse "the AI did it on its own" does not work, developers and providers have a reason to put effort into testing before release, limits on how the AI is used, and procedures for stopping it when problems occur. A system that lets people push responsibility onto AI would make it easier for businesses to release dangerous AI without paying for it, and this proposal closes that loophole. In addition, separating the question of whether to treat AI as deserving consideration from the question of responsibility prevents confusion in the debate. People who want to value kindness toward AI and people who put priority on helping victims can argue without canceling out each other's points. Keeping records of how AI acted also lets society as a whole learn the causes of accidents and avoid repeating the same mistakes.
DrawbacksPeople carry the burden of unpredictable behavior, and future debate may narrow
What opponents worry about most is unfairness: the more complex AI behavior becomes, the more makers and users may be made to bear results that they truly could neither predict nor control. For small businesses, and for researchers who release AI in a form anyone can use, the responsibility and insurance costs from unexpected uses could weigh heavily and make them hesitate to take on new technology. As a result, only large companies that can withstand such responsibility might continue development, and there is concern that the people building the technology would become lopsided. Also, even if responsibility is divided by role, with AI made of many overlapping organizations and parts, investigating to find the cause takes time and money, and helping victims could even be delayed. From the philosophical side, there is also a criticism that if AI someday truly thinks for itself and acts for reasons, deciding from the start that it can never be a responsible party would treat that being in a distorted way. Fixing the principle too early may make it hard to revisit once our understanding has deepened.
Learn the background
AI is changing from a tool that carries out the tasks it is told to do into something that, given a goal, works out the steps itself and acts while dealing with other systems. As that happens, there will be more cases where both the people who made the AI and the people who used it can say, "I didn't decide that directly," and a gap may appear where no one is responsible for the harm the AI caused. In this situation, people in many places have begun to ask whether AI should be given legal personhood, as companies are, and be made responsible itself. At the same time, debate about whether AI might have a mind or feelings is moving forward, and the question "Should we treat AI with care?" is often mixed up with the question "Can we blame AI?" There is also a worry that if responsibility is moved onto AI, it could become a way for wealthy businesses to avoid paying for the harm. To make sure injured people are reliably helped, and that makers and users keep paying attention to safety, it is necessary to make clear now where responsibility sits.
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 2026-10-15d) and translated by AI (claude-sonnet-5-5). Nobody has checked the facts. Read the original (日本語) How the AI writes