By Raphael ‘Tsavkko’ Garcia
When Folha de S.Paulo, Brazil’s largest newspaper, acknowledged that columnist Natalia Beauty had relied extensively on generative artificial intelligence to produce her opinion pieces in February 2026, the controversy quickly moved beyond one writer or one newsroom. It became a test case for journalism itself and, in a sense, anticipated a debate that is taking over the world about AI and authorship.
The immediate question seemed straightforward: how much AI is too much? But the debate that followed exposed a more fundamental uncertainty. If an opinion column can be largely drafted by AI, what exactly makes it the work of the person whose name appears above it? Is authorship defined by typing the words, by generating the ideas, or simply by accepting responsibility for the final text?
What is acceptable when it comes to AI use in the newsroom?
The discussion has divided Brazilian journalists, academics and editors. Some see AI as simply the newest tool in a centuries-old evolution of writing, while others argue that opinion journalism occupies a special place because readers expect not merely competent prose, but a distinctly human perspective.
The question also involves disclosure, and it may soon become partly technical rather than merely editorial. In August, Anthropic announced that new Claude models would embed an imperceptible, machine-readable watermark directly into generated text, a measure introduced as part of its compliance with the EU AI Act’s transparency requirements (even though it goes beyond the original scope and doesn’t take into account the exceptions to the rule).
For journalism, that development adds another layer to the Natalia Beauty controversy: if AI-generated prose can increasingly carry its own detectable provenance, newsrooms may find it harder to treat substantial AI authorship as an entirely private matter between columnist and editor.
But watermarking also complicates the boundary the controversy has exposed as a detectable AI signature could show that a model participated in producing a text; it cannot, by itself, answer the more important editorial question of whether the journalist used AI to polish their own argument or outsourced the intellectual work of constructing it.
In that sense, technology may become increasingly capable of identifying AI’s presence while leaving journalism with the much harder task of deciding what that presence means.
Author or assistant
For veteran journalist and writer Paulo Markun, the controversy should not be reduced to how many sentences were produced by a machine.
“I wouldn’t draw the line based on ‘how much the machine wrote,’ but on another question: who decided what the text wants to say?” he told MDI. “If AI helps organise, test, summarise, rewrite or challenge the author’s arguments, we are in the realm of assistance. If it starts determining the thesis, the criteria of relevance, the hierarchy of facts and the point of view, then it is no longer assistance; it becomes a substitution of the author’s intellectual contribution.”
That distinction, he argues, is more important than counting prompts or measuring percentages of AI-generated text. Writing, after all, has rarely been a solitary activity.
“This dilemma did not begin with AI,” Markun says. “The history of writing has always lived with unstable, collective, attributed, edited or mediated forms of authorship.”
Journalism itself has long relied on editors, researchers, speechwriters, and ghostwriters. AI, in his view, merely makes those long-existing ambiguities impossible to ignore.
Yet Markun rejects the idea that minimal involvement is enough to claim authorship.
“The distinction between authorship and curation lies in the degree of intellectual command,” he says. “The author defines the relevant question, chooses the angle, assumes the risk of interpretation, decides what enters and what stays out, and answers for the public consequences of what is published. The curator selects, organises, approves, improves or validates something produced by someone else.”
The need for transparency
That distinction becomes particularly important when opinion writing is involved.
“If the columnist provides only generic suggestions and performs superficial editing of an AI-generated draft,” he argues, “what remains is little more than the endorsement, the name, the reputation and some curation.”
Even as he defends AI as a legitimate tool, Markun believes transparency is essential.
“If AI produced a complete first draft, structured the main argument or wrote a significant part of the published text, the reader has the right to know,” he says. “The fundamental point is to make it clear that AI is not the author, but that it participated in the production process.”
For technology journalist Cora Rónai, the controversy says less about Natalia Beauty than about the profession’s collective uncertainty.
“I think the debate reveals how unprepared we are to deal with AI,” she says. “I don’t mean that as an accusation, but as a fact: the tool is simply too new, and we’re all learning how to use it.”
That learning process extends well beyond journalism.
“What this story seriously brings us to reflect on,” Ronai says, “is that in many cases AI does the job better than many people.” However, she draws a distinction between technical quality and originality.
“The results are never exceptional,” she says, “but we have to acknowledge that they raise the general standard. AI has raised the floor of writing. It hasn’t necessarily raised the ceiling.”
Redefining authorship?
That observation challenges one of journalism’s traditional assumptions: that the ability to write well is itself evidence of expertise. If competent prose becomes widely accessible, professional value must increasingly come from elsewhere. And for Ronai, authorship is becoming more difficult – not less meaningful.
“The question of authorship is interesting because we’re entering an area that is still undefined,” she says. “For those who write, AI is a tool of expression, and obviously it produces different results depending on who uses it.”
Marun agrees, noting that “the crucial point is not whether a tool was used. It is whether there was genuine intellectual leadership.”
Rather than seeing AI as something fundamentally alien to human creativity, Rónai argues that it extends an existing process, AI “amplifies” our thinking and our style as “they come from what we’ve learned and read throughout life.”
But she does not believe technology reduces editorial responsibility. Quite the opposite. “In a world drowning in dubious information, editors have an even greater commitment to readers.”
That responsibility, she suggests, cannot be outsourced – even if parts of the writing process increasingly can. If Markun and Ronai see AI primarily as a new stage in the evolution of writing, professor of media studies at Rio de Janeiro State University (UERJ), Erick Felinto argues that opinion journalism demands particular caution because its value lies precisely in the expectation of human judgment.
“Traditional journalism is already going through an enormous crisis of credibility,” he says. “This crisis tends to increase if readers discover that an opinion article was largely written by AI.”
Unlike routine reporting, opinion columns are consumed not simply for information, but because readers seek a particular person’s worldview and “if AI can generate competent opinion articles from prompts, textual competence alone ceases to be sufficient.”
Rather than proposing an outright ban on AI, Felinto argues for distinguishing between using AI to support journalism and allowing it to become the primary producer of journalistic expression.
“It seems to me that the clearest principle is to concentrate the use of AI on certain kinds of research rather than on the writing of the texts themselves,” he says. “It is undoubtedly important to make clear to the public which parts or aspects of a report or article made use of AI.”
The challenge, he suggests, is not simply technological but philosophical. AI forces journalism to revisit questions that literary theorists such as Roland Barthes and Michel Foucault raised decades ago about the meaning of authorship itself.
“There is no way to avoid this process, its diffusion will profoundly reconfigure the notion of authorship,” he says.
The need for transparency
Ironically, all three interviewees converge on one point despite their different perspectives: transparency. Readers have the right to know if AI was used to produce an article, editorial responsibility increases and readers should know when AI substantially contributed to a published text.
Ultimately, explains Felinto, “the role of judging and critiquing the use of AI in texts should always fall to the reader. It is possible that, as readers become increasingly familiar with artificial intelligence tools and their peculiarities, a form of balanced critique will emerge regarding the appropriate uses of these technological capabilities.”
That emerging consensus mirrors a broader regulatory shift. From 2 August 2026, Article 50 of the EU AI Act introduces new transparency obligations for AI-generated content.
Among other provisions, it requires disclosure when AI-generated or AI-manipulated text is published to inform the public on matters of public interest – unless the content has undergone meaningful human editorial review and a natural or legal person assumes editorial responsibility. The regulation does not prohibit AI-assisted journalism; instead, it seeks to preserve transparency and accountability.
That distinction is striking in light of the Folha controversy. The newspaper’s debate was never really about whether AI should be used – few journalists reject its use altogether. Rather, it was about whether readers deserve to know when AI has moved beyond spellchecking, research assistance or stylistic editing to become a substantive participant in producing a signed opinion piece.
The real test of authorship, Markun argues, is not whether someone physically wrote every sentence, but whether they exercised genuine intellectual direction.
He cautions against portraying the debate as one between flawless humans and unreliable machines. “Opinion is something AI also produces, and on a massive scale,” Markun says. “Human value lies in something more demanding: doubting one’s own inferences, resisting the pressure of the majority, verifying before condemning, recognising when a convenient narrative is becoming an injustice – and correcting oneself before reality forces that correction.”
The future of journalism
Ultimately, the Natalia Beauty case may be remembered less as a scandal over one columnist than as the first high-profile confrontation with a question every newsroom will soon have to answer.
Journalism has spent decades asking whether machines could write like humans. That question is increasingly being answered in the affirmative. The more difficult question is what readers expect when they see a byline.
If the value of journalism has never simply been arranging words into grammatical sentences, then AI does not eliminate the need for journalists. It raises the bar for what journalists must uniquely contribute: judgment, accountability, originality and the willingness to stand publicly behind an interpretation of the world.
The future of opinion journalism may therefore depend less on whether AI writes the first draft than on whether the person signing the column still owns the thinking behind it.
The views and opinions expressed in this article are solely those of the author and do not reflect the official policy or position of the Media Diversity Institute. Any question or comment should be addressed to [email protected]
