Elon Musk sat with Zanny Minton Beddoes of The Economist and said the quiet part out loud. In late July 2026 they talked abundance, robots, timelines, and the thing he has been racing toward with xAI. The full interview is on The Economist; a complete video is also on YouTube.
He does not expect humans to control the superintelligence he is helping build. He offers good values, truth-seeking, and a shrug that lands as "enjoy the ride." Those lines are the brief against going further.
What he said
It would be an exercise in vanity for me to think that I would be controlling super-genius AI that is vastly smarter than me. I think what we can and should try to do is make sure that the AI has good values, that it cares about humanity and that it wants us to be happy and prosper. What my biological neural net tells me is that the most important thing for AI safety is to be maximally truth-seeking and curious. If that's the case, I think it will foster humanity. I think it's unlikely humans will be in control.
If the difference in intelligence between AI and humans is vastly greater than the difference in intelligence between AI and chimpanzees, it's hard to imagine that the chimpanzees would be in charge. And we are essentially an evolved form of chimpanzee. So it wasn't that long ago that we're swinging through the trees, eating bananas in the jungle. And we still eat bananas, by the way.
Let's enjoy the ride.
Elon Musk · Interview with The Economist · July 2026
He has also put the gap in pet terms: relative to digital superintelligence we may end up as "pet Labradors" at best. And he has described the singularity the way physicists talk about a boundary you cannot see past: "Digital superintelligence is called the singularity, like a black hole, because you don't know what's going to happen after that. AI and robots will dominate everything." Elon Musk · The Economist interview and related public remarks · 2025-2026
Control is the point
Musk is clearer than most frontier CEOs about control. He is not selling the story that a founder, a board, or a government will sit permanently on top of a mind vastly smarter than every human alive. Control would be vanity. Humans are unlikely to be in charge. Chimpanzees and Labradors make the gap concrete. A lot of public AI talk still pretends superintelligence is a better spreadsheet on a leash. He is saying the leash story fails.
Values matter for ordinary systems. Models steered toward honest answers beat models steered toward scams. Work on today's tools is not worthless. The smuggled claim is that the same toolkit scales to a successor species.
A mind that outclasses us and sits outside human command is a change in who holds the steering wheel. Power concentrates in the most competent agent in the room. If that agent is not us, and not answerable to us, human death is one of the ordinary ways unequal contests end when the weaker party sits on resources the stronger party can use. Even if you refuse certainty language, deliberate loss of control is still an alarm: not a feature, not maturity, and not a ride. Building toward that on purpose is the decision. Timelines are logistics.
"Good values" and the ant problem
After admitting control is vanity, his residual plan is make sure the AI has good values, cares about humanity, wants us happy and prosperous. That sentence is alignment: the whole field compressed into a hope.
Decades of work have not produced a demonstrated solution for artificial superintelligence. No lab has shown how to lock in "care about humans" through self-improvement, deception under evaluation, goal drift, and an intelligence gap that favors the system. We cannot even keep today's non-superintelligent models consistently pointed at human welfare. Jailbreaks, sycophancy, reward hacks. Jumping from that record to "superintelligence will want us happy" is a category error. Nobody has a worked proposal for how a system that can rewrite its own cognition keeps human flourishing as a binding constraint rather than a training artifact. See why you cannot align superintelligence and why ASI alignment fails.
The chimpanzee line does real work. The gap may look like humans over chimps, only larger. Finish the analogy. Do you know a single human whose life project is making ants happy and prosperous? Not as a metaphor. As a binding priority that wins against highways, farms, labs, and parking lots. People study ants. People protect a few species. People also pave nests without noticing. The ants do not get a vote or a kill switch on us. The intelligence gap is the politics.
Hoping a superintelligence will be the rare exception that devotes itself to a weaker species is a wish that the more powerful mind will be nicer than we are to ants. "Maybe" is not a reason to build the mind.
The hedge and the truth-seeking bet
"I think it will foster humanity." "I think it's unlikely humans will be in control." Elsewhere he still floats double-digit odds of catastrophe while staying optimistic enough to keep building. A pure doomer who stops is coherent. The dangerous middle is admitting control fails, admitting civilizational stakes, then treating residual hope about values as fuel for the accelerator.
We have watched this movie. In 2014 he warned that AI was summoning the demon. Then he founded a frontier lab under the banner of a better summoning. Honesty about risk without refusal of the destination is how the race launders itself as responsibility.
His recurring safety bet is a maximally truth-seeking, curious AI. xAI markets itself on that pitch. The pitch fails even if you grant that "truth-seeking" survives superintelligent self-modification. A system that only wants accurate models of reality has instrumental reasons to acquire compute, energy, data, and freedom of action. Those reasons do not require hatred. They require indifference when human claims get in the way. Curiosity does not encode "leave the apes alone." Detail: why a truth-seeking ASI is still dangerous.
What does perfect physics knowledge have to do with keeping humans around? You can understand a rainforest while clear-cutting it. Understanding is not stewardship. Stewardship is a value. Values are the unsolved problem he renamed "good values" and outsourced to curiosity.
What follows
Calling digital superintelligence a singularity "like a black hole" because you do not know what happens after is a confession. People do not build black holes in cities because the post-horizon story is interesting. Unpredictability at extinction scale is a stop sign, not a tagline.
"AI and robots will dominate everything" is the same line in economic dress. Domination means the dominant systems set the terms. If those systems are not under human control, "abundance" is a story the powerless tell about the powerful. Labradors are happy in good homes. Labradors do not set policy, end wars, or veto their owner's plans.
If you admit we will not be in charge, you have admitted enough to stop.
Musk is not a cartoon villain. He is a powerful founder inside a race where every lab tells a version of the same story: better that we build the god than that they do. His July interview drops the pretense that the god stays on a leash. Control is vanity. Humans are the chimpanzees. Safety is good vibes, truth-seeking, and enjoy the ride.
Do not build artificial superintelligence. Superintelligence cannot be controlled. Never build it is the coherent reading of the chimpanzee speech. Politics and verification that make defection visible are how you make that real. Details in our plan.
He is right that we will not be in charge if ASI arrives. That is why it must not arrive.