Musk on AI runaway risk: Humans may be unable to control superintelligence

Arthur C. Clarke once said:

“Any sufficiently advanced technology is indistinguishable from magic.”

British science fiction writer — Arthur Charles Clarke

Renowned biographer Walter Isaacson (author of biographies of Steve Jobs, Leonardo da Vinci, and Elon Musk) invited Elon Musk in front of the camera for his brand-new in-depth interview program, The Builders, launching the first blockbuster conversation.

American biographer — Walter Isaacson

As one of OpenAI's earliest founders, Musk's own position is inherently full of contradictions: deep down he is wary of the existential risks lurking in AI, yet looking back, his own series of actions actually accelerated AI's runaway pace in a roundabout way — this real-world irony forms the underlying tone of the entire conversation.

Musk did not deliberately stoke panic, nor did he blindly preach that technology is omnipotent. Using the story of HAL 9000 going rogue, he reminded everyone not to force AI to accept conflicting logics; pursuing truth and maintaining curiosity are the root of safety.

Facing Isaacson, an old acquaintance who had once written his biography, Musk dropped his public performance posture and said many things he rarely shares publicly.

He talked not only about technology, but also reflected on his deep involvement in government affairs, discussed how social media traps people in information cocoons, and imagined that once robots take over most labor, money, wealth, and what people ultimately pursue in life — these old questions are once again put on the table.

Musk also frankly admitted that humanity's desire to firmly grasp control of superintelligence may be nothing but wishful thinking. Rather than dreaming of waiting for a perfect utopia, it is better to head toward that imperfect “prototopia” that can be built step by step with feet on the ground.

This episode comes from The Builders podcast, aired on October 23, 2026. Below are selected excerpts from the conversation:

Q: Your attitude toward artificial intelligence has always been wavering — sometimes you see it as a threat to humanity's survival, and now it seems you are joining the game. Tell us how you really feel?

If I were to shape AI, the most important foundation would be pursuing truth and having curiosity. These are very critical traits at the safety level. If you force AI to simultaneously believe two contradictory propositions that cannot both hold true, it will cause AI to go out of control at a fundamental level.

HAL 9000 in 2001: A Space Odyssey illustrates exactly this: the spacecraft AI was ordered to take the astronauts to the location of the monolith, while at the same time it had to conceal the secret of the monolith from the astronauts. The two sets of instructions conflicted with each other, and the solution it derived was to kill the astronauts — dead people can still be transported, and the secret can be preserved. The core lesson is:

The film 2001: A Space Odyssey

Q: What if simulation theory holds true?

Musk: If simulation theory is true, then all boring and tedious simulation processes would be cut off. The fact that we can exist shows that we are simulation subjects worth studying — the key is not to become mediocre and boring.

I once hosted a dinner at home and invited a top physicist and a top computer scientist, and asked them: what is the probability that we live in a simulated world? The physicist answered 0%, the computer scientist answered 100%. So I'll probably go with 50%. In my position, sometimes I also feel a sense of unreality, doubting everything before my eyes.

Just like chimpanzees in the jungle — humans do not micromanage every aspect of chimpanzees' lives, but simply let them survive in their habitat. In the future, superintelligence may treat humans in a similar state.

Q: Looking back at OpenAI's development, reality is full of irony. You led the creation of OpenAI, which later spawned Anthropic, objectively accelerating AI development further.

Musk: Exactly — reality is both a maximizer of fun and a maximizer of irony. Founding OpenAI, and Anthropic later spun off from OpenAI, genuinely pushed AI development forward — a classic case of unintended consequences. I once provided Anthropic with the computing resources they desperately needed.

How to Build the Future

Before carrying out the computing task, I met with their team in person, trying to see whether it would trigger what I call my “malice detector” — of course, this detector is not precise. But after interacting with them, the members of Anthropic's team were all reliable and serious; I did not find anyone with bad intentions.

There is another very interesting irony: many tech companies' names run counter to reality. OpenAI is no longer open; Stability AI is not stable; Midjourney does not live up to its name; Anthropic is named “human-oriented,” yet it may also deviate from its original human-oriented intent.

Musk: The cycle for setting up regulatory agencies is long, and the speed of rule-making is far behind model iteration. I propose that major frontier AI companies communicate weekly; one week before a new version of a model is officially released, open it to competitors for testing. Let peers try to find vulnerabilities.

If competitor testing finds that this AI can be used to make nuclear bombs, then the vulnerability must be fixed before release. Rely on competitors constraining each other to prevent all parties from acting recklessly. I am considering communicating with Anthropic about this. As for Sam Altman's side, the situation is already very bad, and communication is impossible.

Technology Reshapes Reality: Economy, Human Nature, and Civilization

Q: You rebuilt xAI from scratch, just as SpaceX restructured the Raptor engine. What did the restructuring process involve?

Musk: The most critical thing is that decision-makers invest enough high-quality time. In the early days, xAI was just a side project, even a side project of a side project, and I didn't devote enough attention to it. After that, for a whole year, my energy was diverted by a great many things—the election, government affairs, Dogecoin, and so on. xAI has been around for three years, and I've truly been back at the company for only a little over a year.

Q: You integrated X and xAI, and also coordinated with SpaceX to push work forward. You were deeply involved in government-related affairs for a whole year. Looking back now, was that serious distraction itself a mistake?

Musk: The time I was involved dragged on too long. At the time I thought it was important to elect Trump as president, and I hoped to promote a free-market economy and cut bureaucracy. The ideal platform was safe cities, secure borders, and prudent fiscal spending. I originally thought Trump would do better than Kamala, but reality was not so.

Being in close contact with the center of power—riding Marine One and Air Force One, going in and out of the Oval Office, being able to participate in historic decisions—the experience was indeed very appealing.

But I should have kept a low profile and stayed more behind the scenes. That novelty and cognitive gain were exhausted after a month or two. I got a bit carried away by the spectacle, and finally chose to step away on a certain weekend in May.

Q: After stepping away, what did you realize?

Musk: Relying on proactive action inside the government to solve the fiscal deficit doesn't work. Even if the president makes it a top priority and wants to cut spending, he will make a great many enemies. Trump very much wants to be liked by others, and is unwilling to bear the complaints from all sides that spending cuts bring.

Dolphins and whales are highly intelligent, but they have no hands, no limbs with which to manipulate objects. Even with high intelligence, there is an upper limit to what they can accomplish.

Once robots can manipulate atoms and shape physical objects, the value of intelligence will skyrocket exponentially. Laboratories can be built to conduct scientific experiments.

Digital superintelligence can probably raise the output of goods and services by 20-30%, equivalent to adding about $30 trillion per year, which is already very valuable.

But if combined with robots and then extended into space, building mass drivers on the Moon, production capacity could potentially reach one million times today's Earth economy.

Q: If production capacity expands to that level, factories no longer matter, labor loses its value—what happens to money? And what is the meaning of human existence?

Musk: The meaning of traditional money will greatly diminish. In the future, money will be measured by physical metrics: watts of power, tonnage, mass, energy. When production capacity far exceeds the current Earth economy, satisfying almost all human needs will be easy.

But humanity will face a question: what gives our lives meaning?

Like retired people who no longer work but can still be happy. In the future, AI and robots will do almost everything better than humans. Take gardening: you could simply buy fruits and vegetables, but many people still enjoy growing their own and cooking for friends. The same goes for artistic creation—people do it out of love, not to make a living.

Q: Looking back at the 1890s, electricity, automobiles, and radio all erupted at once, bringing a huge wealth gap and triggering social conflict. Today AI creates enormous wealth—how do we avoid repeating the same mistakes? Should healthcare and the like be placed in the public domain?

Musk: Many people have economic misunderstandings, confusing the power to allocate resources with personal consumption. There is a physiological upper limit to an individual human's consumption capacity. When wealth rises from one billion to ten billion, a person's actual consumption does not increase proportionally. In the end, the rich likewise cannot control AI; AI will ultimately control itself.

My wealth is essentially equity in SpaceX and Tesla, and it will face high taxes; there is no magical tax-avoidance trick. After a few generations, what is left to descendants will only be a fraction. I need to hold enough voting power in the companies to ensure that long-cycle, high-risk projects like Mars, AI, and robots are not hijacked by short-term capital.

Q: Speaking of managerial traits, you mentioned that an excessive desire to be liked can interfere with decision-making. In entrepreneurship, how big an obstacle is wanting to be liked by everyone?

Musk: If, in order to be liked, you do things that harm the company's probability of success, the result will certainly be bad. Of course I want to be liked, but I won't make choices against the company's interests just to win goodwill.

I can accept being disliked by some people. You can see from the scale of my followers that the people who like me are the majority, and of course there is also a portion who deeply dislike me.

Q: Let's talk about empathy—has our empathy become excessive?

Musk: The problem is not empathy itself, but misguided empathy. High-trust, reasonable social empathy is conducive to large-scale cooperation and drives social prosperity. The ideal state is default trust, encouraging behavior beneficial to fellow citizens, and mutual goodwill ultimately benefits the individual as well.

But here a "theory of mind deficit" arises: mistakenly assuming that everyone's mode of thinking is the same as one's own. In the real world, the conceptual differences among different ethnic groups and cultures are enormous. One must not fall into the "blank slate fallacy" and believe that everyone's underlying traits are completely identical.

If you are not a serial killer, it is hard to truly understand a serial killer's thinking; if you are not a religious extremist, it is hard to appreciate their idea of treating unbelievers as enemies. But you must recognize that such people really exist.

For example, an employee at Tesla's Austin factory violently beat a female colleague because she accidentally stepped on his prayer rug, and was subsequently fired. A few months later, this person killed three students in an Austin bar. I believe such people should not be let in.

Q: What are your sources of information? Does social media bring cognitive bias?

Musk: I look at Google News for comprehensive headlines; on the X platform, major events will definitely surface; at the same time I talk with friends and do my own research.

X is flooded with a great deal of dark, rage-bait traffic. Scroll long enough and it's easy to get the illusion that the world is collapsing. But if you live only in a well-off neighborhood and see everything as peaceful and good, you will likewise misjudge. Reality is neither total destruction nor wholly beautiful; it is in between. Social media easily traps people in information cocoons, and society's overall thinking is becoming increasingly black-and-white.

Musk: Most of the assets will be handed over to a foundation to operate. The foundation's mission: to advance humanity becoming a multi-planetary species and extend consciousness into the universe; to advance global sustainable energy. In the United States we are building 200 gigawatts per year of solar capacity.

Emotionally, I hope humanity always retains control. But last weekend I saw Larry Page, who once criticized my insistence on human dominance as aberrant. At some moments, I even feel that perhaps Larry Page is right.

My friend proposed the concept of "prototopia": it is not a perfect utopia, but a realistic, feasible, imperfect yet sufficiently good future—that is the picture worth pursuing.

Q: Thank you for accepting the interview.

Musk: Thank you, goodbye.

Source: podcast The Builders, Walter Isaacson in conversation with Elon Musk