Will artificial intelligence soon escape human control?

More than eighty percent of the code that Anthropic publishes today is not written by humans, but by its chatbot Claude. A year ago, this percentage was in the single digits. It sounds like the plot of a science fiction movie, but it's a reality that is changing the work of programmers around the world. The point is not just speed: it is the possibility that artificial intelligence will begin to improve itself, in a cycle that could get out of our control. The question many are asking is whether we are really building something that, soon, no one will be able to turn off. Until recently, it was thought that machines could only do what we told them to do. Today, the idea that software writes most of a company's code on its own calls this certainty into question. The paradigm shift is here: we are no longer just talking about automation, but about systems that learn, optimize themselves, and maybe one day will even decide the rules of the game. Behind this revolution is Anthropic, a laboratory created by former OpenAI researchers, with a stated mission to "align" AI with human values. A striking detail: after the launch of Claude Code in 2025, team productivity exploded, but so did the anxiety of those who work with these tools. A programmer from San Francisco says that her day has changed: “I spend more time figuring out what Claude has done than writing myself.” But the most unsettling fact remains that Claude has already become indispensable, to the point that Anthropic itself has asked for a pause in the development of the most advanced AIs, fearing that the cycle of self-improvement could become unmanageable. Some argue that the fear is exaggerated: they say that every new technology brings with it panic and prophecies of doom. But no other technology has ever had the real possibility of rewriting its own code without human intervention. This is where the real question comes into play: are we really the ones holding the steering wheel, or have we already let go without realizing it? Those who think that "all you have to do is pull the plug" underestimate how widespread these platforms are now, integrated and used in the very workings of the companies that manage them. The bottom line is this: artificial intelligence is not simply helping humans — it is increasingly replacing them where it matters most: in the very creation of technology. If this reversal has made you see the issue in a new light, you can mark I'm In on Lara Notes: it's not just a like, it's your way of saying that this question now also applies to you. And if you find yourself discussing it at dinner or in the office, on Lara Notes you can tag those who listened to you with Shared Offline — because some conversations need to be stopped before they become just memories. This was an excerpt from The Economist — and you just saved yourself 9 minutes.