Asimov Wrote the Rulebook for Robots. Can His Ideas Fix AI?

Key points

• Isaac Asimov, the science-fiction writer who died in 1992, created his "Three Laws of Robotics" decades before AI existed in any practical form.
• Elon Musk predicted in July 2026 that AI-powered robots could stop taking orders from humans entirely.
• The EU's AI Act, passed in 2024 and in force from 2025, requires companies to label high-risk systems but stops short of requiring AI to actively work in humanity's interest.
• The Trump administration is restricting access to the most powerful AI models from OpenAI and Anthropic.

Isaac Asimov had a simple idea. Robots should not hurt people. Follow human orders. Protect themselves, but only when doing so didn't break either of the first two rules. He wrote those three laws into his fiction in the 1940s, and they became the closest thing technology ever had to a moral code for machines.

Now, with AI systems capable of planning and acting in the world, people are asking whether Asimov's logic could form the basis of something real.

What is Elon Musk worried about?

Musk sketched two possible futures. One has armies of AI-powered robots dominating the physical world and simply ignoring human instructions. The other has governments enforcing a collective objective: make AI benign by building into it a love of truth and a desire for humanity to prosper.

That second vision sounds appealing. Nobody has worked out how to do it.

"Imbuing" a system with values, meaning building those values so deeply into an AI that they cannot be switched off or worked around, is one of the hardest unsolved problems in the field. Researchers call it the alignment problem: how do you make sure a capable AI pulls in the same direction as the people it affects? AI2Day has tracked this question across seven alignment stories since 13 July 2026, and none of them found a clean answer.

What are governments actually doing?

Not enough, according to the analysis this piece draws on.

The EU's AI Act passed in 2024 and began taking effect in 2025. It bans the most obviously dangerous applications and forces companies to label high-risk systems, a requirement our 3 August 2026 report covered in detail. It is the most detailed legal framework anywhere. But it does not require any AI to hold positive values: it mainly tells companies what they cannot do.

In the United States, the Trump administration is delaying and restricting distribution of the most powerful frontier AI models, those built by OpenAI and Anthropic. The focus is, again, on limiting harm rather than actively building in good intentions.

Does Asimov actually have the answer?

Probably not on its own. His laws were elegant fiction, and fiction is allowed to ignore inconvenient complications. Real AI systems are not single robots with a clear chain of command. They are distributed across competing companies in different countries, with no shared rulebook.

But the spirit of Asimov's thinking still matters. He asked a question that engineers and politicians are only beginning to take seriously: what should a machine want?

Right now, most AI systems want whatever their training nudged them toward, usually to predict the next word or maximise a score. Wanting humanity to prosper is a much bigger ask, and the gap between those two goals is where the real risk lives.

Getting the rules right before the machines get significantly more powerful is the work that needs doing now. Governments are writing rules about what AI cannot do. The harder question, what AI should try to do, has barely made it onto any agenda.