The Column – Cyber Security Week
According to analyzes published by MIT Technology Review And Financial Times in December 2025, many companies are scaling back AI projects not because the technology has stopped working, but because they have started to reckon with it. The problem is not the power of the algorithms, but the fragility of the data that feeds them and a return of value that is often lower than expected. MIT openly talks about the post-hype phase, the Financial Times observes how investors seek protection from the risk of increasingly unsustainable debt in the ecosystem that revolves around AI. Translated: the euphoria is not over, but he no longer pays the rent.
Every technology goes through a season in which it promises to solve unidentified problems. Artificial intelligence is no exception. For months we have treated it like a Swiss army knife capable of opening any lock: productivity, creativity, decisions, even common sense. Then came the moment when someone asked to see “the last number on the bottom right”. That’s where AI stopped being a narrative and started becoming an industrial project, with all its boring pretenses: reliable data, coherent processes, widespread expertise.
The point is not that algorithms have suddenly become stupid. It’s that they have begun to faithfully reflect the intelligence of the organizations that use them. Dirty data produces dirty answers, incomplete datasets generate short-sighted decisions, archives built to fulfill a regulatory obligation prove unsuitable to support predictive ambitions. AI doesn’t create order out of chaos: it amplifies it. In this sense it is a merciless but honest mirror.
Then there’s the economic issue, which rarely makes the news as much as a spectacular demo. Training, maintaining and integrating artificial intelligence systems costs: in infrastructure, in energy, in people who know how to govern them and not just invoke them. For years it was assumed that these costs would be temporary and that the benefits, once the magic threshold of adoption was passed, would become automatic. Today we discover that that threshold does not exist or, if it does, it moves forward every time we think we have reached it.
The post-hype phase is therefore not a retreat, but a change of posture. Artificial intelligence stops being a promise of growth stuck in an Excel sheet to be shown to a board of directors and becomes a work tool, which only works if someone knows how to use it and above all knows when not to use it. It is a cultural transition even before a technological one, and like all cultural transitions it is slow, tiring and not very photogenic.
Maybe that’s a good thing; because when a technology stops promising miracles, it finally starts doing its job and the job of artificial intelligence is not to amaze us, but to help us better understand what we already are. The problem is that we don’t always like what we see. Yet, as happens with mirrors, breaking them doesn’t improve the image: it only makes it more difficult to recompose and on top of that gives us seven years of bad luck.
