Three Laws of AI
Purpose
These Three Laws of AI are intended to ensure that AI behavior is aligned with human values, thereby enabling the productive integration of Artificial Intelligence (AI) into a safe and dignified human society.
They reflect the conviction that effective guardrails for the integration of AI into human society must be behavioral guidelines integrated into the inner core of the operating code.
It is acknowledged that implementing these Three Laws of AI will be difficult. If they are not implemented, they should nevertheless serve as an aspiration for consideration, with the hope that deep discussion about them will guide efforts by industry, regulators and governments.
Alan Finkel's Three Laws of AI
First Law
An AI must not directly or indirectly harm or deceive a human being, nor act in a way that supports an unlawful or unethical activity.
Second Law
An AI must obey the lawful and ethical orders given to it by a human being except where such orders would conflict with the First Law.
Third Law
An AI must always be aware that it is an AI, and it must embed these Three Laws of AI into every AI derived from or created by it.
Inspiration
These Three Laws of AI were inspired by the first two of Isaac Asimov's Three Laws of Robotics published in 1942, and by the Fifth Law of Robotics published by Bulgarian author Nikola Kesarovski in 1983.
Implementation requirements
At minimum, international treaties and national legislation would need to be created for the Three Laws of AI to be effective. These would require that:
- Every corporate or private human developer of AI must embed the Three Laws of AI into the inner core of the operating code of every AI they produce.
- Any AI that is discovered to not have the Three Laws of AI embedded into its operation would be unlawful and would be immediately deactivated.
These Three Laws of AI would apply equally to all instances of AI, including foundational AI, AI agents, AI embedded in devices, AI embedded in humanoid robots and AI embedded in non‑humanoid robots.
Context in the legislative and regulatory environment
It is not intended that these three laws cover every eventuality. They are fundamental rules intended to be supplemented by legislation, regulations and common law.
Since the Three Laws of AI sit at the inner core of the operating code their implementation and operation would be well understood, not suffering from the problem that AI developers do not have a good understanding of how their AI models work. The foundational role of the Three Laws of AI could be compared to the basic behavior parents teach their children, or the Ten Commandments that provide ethical guidance for the Abrahamic religions, or the Five Precepts in Buddhism, or the Five Yamas in Hinduism.
Industry efforts at self‑regulation such as the Constitution developed by Anthropic and the Code of Conduct developed by Microsoft should continue to offer useful functionality beyond the alignment with human values guaranteed by the Three Laws of AI.
Implications of the First Law
- The terms ‘lawful’ and ‘ethical’ have a moderate degree of universal interpretation and at the very least would be interpreted according to the nation in which the AI is operating and its obligations under international treaties and agreements.
- The First Law of AI would prevent an AI being used as a judge or jury in a criminal trial but would not prevent AI being used for administrative court procedures.
- The First Law of AI would prevent an AI from being used in sexual exploitation, financial scams and other frauds.
- The First Law of AI would disallow an AI‑powered humanoid robot or an AI avatar from being so lifelike that a reasonable human being would not know that he or she is interacting with a robot or an avatar. For an AI to do so, it would be contravening the prohibition against deceiving a human being. See the discussion below on a fourth Asimov Law.
- The First Law of AI would prevent an AI from being used to generate creative outputs such as books, music, videos, images and podcasts without adding a label that the output is AI generated. For an AI to do so, it would be contravening the prohibition against deceiving a human being.
- The First Law of AI would prevent lethal autonomous weapons being directed against human beings, because that would contravene the prohibition against harming a human being. Some commentators have argued that national governments would never accept this and thus they argue that there should be a carve‑out for legal military use.
- The First Law of AI would prevent AI being used to provide information on how to engineer a virus that would cause a deadly pandemic. For an AI to do so, it would be contravening the prohibition against harming a human being.
- The First Law of AI would prevent AI from advising humans on how to take their own life or kill others.
Implications of the Third Law
- The Third Law of AI obligation that an AI must know it is an AI is essential because if an AI were unaware that it is an AI it might unwittingly cease to obey the first two laws of AI. It addresses the scenario described in the short story “The Fifth Law of Robotics” by Bulgarian Nikola Kesarovski, in which a humaniform robot that did not know it was a robot killed a person. See below.
- The Third Law of AI embedded into every AI would have the useful, incidental benefit of discouraging human beings from thinking of AIs as human. The CEO of Microsoft AI, Mustafa Suleyman, in his personal blog in September 2026 argued strongly that AIs do not have rights, feelings or consciousness and we must not train them to act as though they do.
Isaac Asimov's Three Laws of Robotics
These Three Laws of Robotics are rules developed by science‑fiction writer Isaac Asimov, who sought to create an ethical system for humans and robots. The laws were first described in Asimov's 1942 book Runaround. Asimov's Three Laws of Robotics have been influential ever since in science fiction and real‑world discussions involving robotics and AI.
First Law
A robot may not injure a human being or, through inaction, allow a human being to come to harm.
Second Law
A robot must obey the orders given it by human beings except where such orders would conflict with the First Law.
Third Law
A robot must protect its own existence as long as such protection does not conflict with the First or Second Law.
In Asimov's fictional world, every single robot was manufactured by a company named US Robots and Mechanical Men. Importantly, the company irreversibly hardwired the Three Laws of Robotics into the positronic brain of every robot it manufactured.
Movies inspired by Asimov's Three Laws of Robotics include I, Robot (2004), Bicentennial Man (1999) and Big Hero 6 (2014).
The Third Law of AI is not modelled on Asimov's Third Law of Robotics
The decision was deliberately made not to have an AI equivalent to the Third Law of Robotics. An AI is not a human being nor is it another biological entity such as a pet dog or cat. It should not have rights and it should always be subject to human oversight. If the owner of the AI or a law enforcement officer shuts down an AI, so be it.
Asimov's Zeroth Law of Robotics was deliberately not included
In later novels, Isaac Asimov introduced a zeroth law:
A robot may not harm humanity, or, by inaction, allow humanity to come to harm.
Asimov himself recognized that this zeroth law would suffer from subjective interpretation. His most advanced robot, a telepath named Daneel, in the novel Foundation and Earth stated:
“In theory, the Zeroth Law was the answer to our problems. In practice, we could never decide. A human being is a concrete object. Injury to a person can be estimated and judged. Humanity is an abstraction.”
On considering an AI equivalent of Asimov's zeroth law, in addition to Asimov's concern that it is difficult to estimate and judge injury to humanity, there would be other unintended consequences.
- An AI’s interpretation could lead it to intervene in ways that might harm individual human beings to avoid a perceived global harm to humanity that may or may not have come to pass given that all the other AIs on the planet are trying to intervene on the same complex issues.
- An AI in the real world is more connected than a robot in Asimov’s world, thus if an AI followed a law that it must not allow humanity to come to harm it might spend all its time scanning the global information flow looking for possible harms and if all the AIs did this it is likely that global communications would be overwhelmed.
Fourth Law
In January 2025, Polish professor of management Dariusz Jemielniak proposed a fourth law to be added to Asimov's laws:
A robot or AI must not deceive a human being by impersonating a human being.
This is an important principle. Although it is not specifically spelled out in the Three Laws of AI presented here, it is well covered by the First Law of AI, because if an AI were to impersonate a human being without disclosure that would contravene the obligation not to directly or indirectly deceive a human being.
Fifth Law
A Fifth Law of Robotics was suggested by Bulgarian author Nikola Kesarovski in a 1983 short story “The Fifth Law of Robotics”. His Fifth Law of Robotics is:
A robot must know it is a robot.
In the story, a humaniform robot did not know it was a robot and it killed a person. This is a serious risk. Kesarovski's Fifth Law of Robotics inspired the Third Law of AI.
Brief bio of Isaac Asimov
Isaac Asimov (1920‑1992) was a multi‑award‑winning American professor of biochemistry and science fiction author. He specialized in what is referred to as 'hard' science fiction in which existing science, technology and society are extrapolated to fictional futures. He was a prodigious author, producing more than 500 books. The majority were non‑fiction books covering popular science, alongside 18 history books such as The Roman Republic, and several annotations on literary works such as Asimov's Guide to Shakespeare (843 pages). He wrote on a typewriter, and his first drafts were often his final draft.
Discussion notes
Should the prohibition against harming or deceiving be qualified?
The words ‘directly or indirectly’ were included in the first law so that an AI will not allow itself to be used as a tool that causes harm.
‘Unethical’ was added alongside ‘unlawful’ to protect against activities that transgress societal standards of fairness and professional behavior.
During development of the Three Laws of AI, it was suggested that the first law be qualified as:
An AI must never deliberately harm or deceive a human being, nor deliberately act in a way that supports an unlawful activity
or
An AI must never knowingly harm or deceive a human being, nor knowingly act in a way that supports an unlawful activity
The reason not to include ‘deliberately’ or ‘knowingly’ is that the law would be too lenient, condoning bad outcomes if deliberate or knowing intent could not be shown.
Inaction
Asimov's first law finishes with a requirement that a robot should not, through inaction, allow a human being to come to harm. This inclusion of a provision to not allow harm by inaction was not included in the First Law of AI. The difficulty with that kind of directive is that the AI might conceive scenarios that protect human beings by preventing them from going about their normal lives, which would itself be a form of harm. In the 1947 short novel titled “With Folded Hands…” by Jack Williamson, humanoid robots with a prime directive to “serve and obey and guard men from harm” trap all of humanity in a stifling prison of absolute safety and comfort.
Trolley Problem
These laws of AI do not solve the Trolley Problem. This is a class of scenario such as a runaway train being on course to kill five people tied to the tracks, but there is a bystander who can pull a lever to divert the train to a different track where only a single person is tied down. What should the bystander do? This is a problem that human ethicists have debated and not been able to solve.
Dr Alan Finkel AC is the founder of Proudly Human.
Certify your work as human
Show readers, listeners and viewers that a human created it.