Artificial intelligence (AI) has arrived in journalism as a promise: fast, inevitable, and transformative. Though imperceptible from the outside, its impact is irreversible and is redefining the speed and scope of journalism from within. At the same time, this fruit of the tree of knowledge comes with structural risks that are not being addressed in public discourse.

These days, AI is not only transforming news production, it is also shaping the very infrastructure of the information ecosystem, from newsrooms to distribution. The result? The media, once proud and quixotic guardians of editorial autonomy, are beginning to look like chain smokers, perfectly aware of their dependence on a system that can destroy them, but unable to kick the habit. Or willing submission.

The danger is not only technological, but political, ethical, and existential. We must ask ourselves whether we are witnessing, as suggested by Felix Simon, a researcher at the Reuters Institute for the Study of Journalism, a “capture of infrastructure” that threatens the very foundations of independent journalism. Or, as Peter Loge, director of the School of Media and Public Affairs (SMPA) at The George Washington University, the “assisted suicide” of newsrooms, caught between algorithmic efficiency and the abandonment of their mission.

Captures, dependencies, vassalage

In his 2023 article “Escape me if you can: How AI reshapes news organizations’ dependency on platform companies,” Simon introduces the concept of “infrastructure capture” to describe the phenomenon whereby private companies—technology platforms—not only provide tools to the media, but also design and control the environments in which journalistic content is created, circulated, and monetized.

It is not simply a matter of the media using third-party AI, but rather that they operate within closed systems (owned by Amazon, Google, OpenAI, or Microsoft) that set the rules of the game. According to Simon, “whether this will have a significant impact on editorial independence remains to be seen,” but everything points to the path toward such capture already being paved.

The perverse part is that this capture need not be coercive. All it takes are so-called “lock-in effects”—increasing dependency whereby, once an organization integrates a foreign infrastructure, the costs of abandoning it become prohibitive. Migrating to another system or building one of your own is, in practice, unfeasible. Like frogs boiled over low heat, the media barely notice how their room for maneuver diminishes with each step.

Simon argues that this situation turns the media into “technological vassals,” trapped in relationships of dependency that sooner or later compromise their ability to decide what, how, and for whom they report.

Beyond direct capture, Simon draws attention to the indirect effects that AI has on the news industry. It is not that a platform decides to censor local media or uncomfortable investigative reports; that would be too crude, almost a nineteenth-century scenario. The real threat is more elegant and therefore more lethal: that journalism will end up competing on an uneven playing field against a barrage of automatically generated content that pretends to be news.

In their own words: “The growing use of AI on digital platforms and the potential risks to the visibility of journalistic content could lead to a progressive hollowing out of the news industry in general.” Particularly vulnerable in this context is local journalism, already hit hard by the economic crisis and the advertising exodus to digital platforms. And, of course, precarious journalists.

Thus, while the mainstream media and journalists struggle to survive in a market saturated with AI-generated fake news, independent journalism, an uncomfortable witness to corruption, injustice, and inequality, to giants and windmills, risks becoming an archaeological rarity, relevant only to nostalgic academics and the occasional collector.

The mirage of efficiency

Peter Loge offers another equally crucial perspective: the mirage of efficiency. AI promises speed, volume, and cost reduction, irresistible temptations for executives who think in terms of quarterly results. However, Loge warns that “newsrooms may be seduced by efficiency and forget that their purpose is not to maximize clicks, but to serve the public interest.”

It’s easy to fall into the trap. If producing ten times more content costs half as much, what executive is going to stop and consider whether that content has value? Why ask whether it maintains the rigor, depth, or social responsibility that have historically defined journalism?

As Simon puts it, “Whether an organization acts in accordance with journalistic standards has little to do with technology and everything to do with how it conceives its mission.”

The disturbing reality is, therefore—as always—that AI or any technological advance, whether the printing press or the internet, does not in itself degrade journalism, but rather it is those responsible for the media who decide, consciously or unconsciously, to sacrifice quality for volume, accuracy for immediacy, truth for profitability.

But the machine poses an added danger. The impact of AI is not limited to content generation. It also reconfigures distribution channels: what news the public sees, in what order, with what headlines, with what priorities. Simon points out that this algorithmic control poses serious risks of bias and accidental misinformation.

AI can amplify erroneous content without any explicit intention to deceive; all it takes is poor model training or poor optimization of the recommendation algorithm.

The danger is that, in an environment where visibility increasingly depends on opaque formulas, journalistic quality criteria will be buried under metrics such as clicks, reading times, or engagement.

To counteract this drift, Simon recommends that media outlets adopt safeguards such as rigorous human verification of any AI-generated or assisted content, systematic testing before implementing automated systems, and active collaboration with universities and other media outlets to share best practices.

In a world where algorithms write headlines, draft summaries, and optimize virality, the only real defense of journalism is ethics. “No one directly forces publishers to use AI or decide exactly how they should use it,” Simon reminds us. The decision, in the end, is human. And deeply political.

AI opens up enormous possibilities for investigating corruption more thoroughly, analyzing large databases, and improving the accessibility of news. At the same time, it can be used to regurgitate low-quality content that simulates depth. We cannot blame AI for this, but rather humans.

And what about the role of governments? Don’t expect a cavalry charge. When it comes to state intervention, we cannot place too much trust in magic solutions from governments. And in the current populist drift, it is better that they do not intervene. Neither Simon nor Loge comment on this.

In any case, skepticism is well founded: how can a political power that barely understands technology legislate effectively on it?

The great codependency

Far from being a simple instrumental collaboration, the relationship between AI and journalism is shaping up to be a toxic codependency. The media need the tools that technology offers to survive in a fiercely competitive market. But at the same time, every step they take toward that dependence erodes their autonomy, their credibility, and ultimately their purpose of existence.

Beyond the internal use of artificial intelligence tools, traditional media outlets are embracing an even more radical trend: selling their content directly to AI platforms to feed their language models. Companies such as OpenAI, Mistral, and Meta have signed agreements with groups such as Reuters, The Guardian, AFP, Prisa Media, and Schibsted. The logic is understandable: new avenues of monetization, defense of copyright against indiscriminate use, and enhanced visibility. At the same time, AI itself can provide the technical solutions for verifying news content. These two developments deserve more detailed and comprehensive analysis.

The dilemma is simple and brutal. Either journalism tames AI to serve its ethical purpose, or AI will tame journalism to serve its commercial purpose. The love of truth should unite man and machine.

AI, let us remember, is not epistemologically competent. It does not know what is true or false: it only predicts what, statistically, should sound plausible. As Felix Simon warns, “the risk of factually inaccurate information appearing in news content can reasonably be controlled,” but only if humans impose strict limits. In other words, AI cannot distinguish truth from a plausible photocopy.

Worse still, in an environment where the volume of content is multiplying exponentially, how can we distinguish real news from the avalanche of plausibilities designed to distract us? If the visibility of information is left in the hands of algorithms optimized to maximize engagement rather than safeguard truthfulness, truth risks becoming a luxury, a secondary concern at best.

Peter Loge, for his part, reminds us that the journalistic commitment is not to efficiency but to the public interest, which presupposes an active search for the truth, not its simulation. AI can be an ally in this task, for example, by analyzing large databases and uncovering hidden patterns, but it can also be the unwitting accomplice of mass trivialization if clear limits are not established.

Perhaps the duty of journalism in the age of AI is to distrust not only algorithms, but also the fascination they exert. Because if the goal ceases to be the truth and becomes simply the optimized production of viral content, what will die is not the news business, but the very idea of journalism.

Journalism must remember, with a Hippocratic oath, that its purpose is not to sound true, but to be true.