A Wall Street Journal opinion piece by investor Stanley Druckenmiller has set off a media industry fight over AI authorship, disclosure and whether a byline still means what readers think it means.
The piece, titled Let the Bond Market Speak, criticized expanded US Treasury buybacks by Treasury Secretary Scott Bessent, who the source article describes as Druckenmiller’s former protégé. After publication, readers said the essay carried a recognizable AI tone and ran it through Pangram, an AI-detection tool that labeled the article 100% AI-generated.
Druckenmiller said he used AI and was not apologetic
What followed stood out because Druckenmiller did not apologize and did not deny using the technology.
Speaking to NOTUS, he said: “Of course I used AI. I’m not embarrassed. I use AI for everything I write now, for the same reason I use a calculator when I do math.”
That response shifted the discussion away from detection alone. The question became whether readers care if an article is AI-written, and whether a signed author must disclose that process at all.
Paul Gigot drew a line between staff writers and outside contributors
Wall Street Journal editorial page editor Paul Gigot did not rebuke Druckenmiller. Instead, he wrote a column mocking what the source article called media elites for objecting to AI-written prose.
Gigot said the Journal’s in-house editorial writers still have to write their own pieces. Outside contributors are different. He said he does not plan to police whether they use AI.
He framed the matter this way: many contributors, especially prominent politicians and chief executives, already rely on speechwriters or outsource drafts to Washington writing shops. In his view, using AI to smooth an article is not clearly different from those existing practices.
Under that approach, the Journal is not treating AI assistance by outside opinion contributors as something that automatically requires separate disclosure.
Earlier comments inside the Journal resurfaced
The episode also drew attention to earlier writing from within the same paper.
Four months earlier, Wall Street Journal editorial features editor James Taranto had described AI-detection tools such as Pangram as “defamation machines.” In that column, he wrote that people who treat writing as a professional craft have an ethical duty, and sometimes a contractual duty, to make sure their work is original.
After the Druckenmiller episode, Taranto kept that position on originality and said Gigot’s current view was consistent with what he had written before. The source article says its author read both positions twice and still could not see how they matched.
The source article tied the dispute to Roland Barthes
The report placed the controversy next to Roland Barthes’ 1967 argument about the “death of the author.” Barthes did not mean that writers disappear. His point was that readers too often treat a text as inseparable from the author’s biography, intentions and identity. Once a work exists, it no longer belongs only to the person who produced it.
That idea now looks less abstract. AI has turned an old literary argument into a practical newsroom problem. It is no longer only about who an author is. It is also about whether anyone can still tell what “written by” means in a stable, operational sense.
“AI is not a calculator, it is a reasoning machine”
The source article rejects Druckenmiller’s calculator analogy. A calculator does not decide what problem to solve. It performs the computation so the user does not have to do it by hand or risk arithmetic mistakes.
Writing an argument works differently. The act is not limited to expressing a conclusion someone already holds. Part of writing is discovering what the argument actually is. A writer may begin by trying to defend A, then realize by the third paragraph that A does not hold up and that B is the more interesting point. If AI drives that process, the article argues, A may remain weak and B may never appear.
University of Maryland researchers call it “argument collapse”
The piece cites a group of University of Maryland researchers who are studying the issue under the label “argument collapse.” In a paper that has not yet passed peer review, they found that AI-written essays are less likely than fully human-written essays to produce novel arguments.
Jenna Russell, a doctoral student studying AI writing in newspapers, put the concern more bluntly: “These large language models will begin to shape our ideology, and we won’t realize it is happening.”
The article notes that this kind of concern is not unique to AI text generation. Social platforms have also long faced criticism for influencing how people think without making that influence obvious in daily use.
Private training data could erase detectable AI traces
The report does not argue for rejecting AI writing outright. Instead, it says people write differently because each person carries a separate reservoir of experience, reading, memory and judgment. What someone has read, what they have suffered, losses taken in a bear market and intimate personal history all function, in the article’s framing, like training data. What that person consistently treats as important works like weighting.
From that angle, AI is not a foreign species. It is a reconstructed knowledge structure modeled on human reasoning.
The article then argues that Pangram was able to catch the Druckenmiller essay because current general-purpose models still show recognizable syntax habits and a shared public average derived from massive common corpora. Readers notice that pattern as the “AI feel.”
That may not last. As personalized AI develops, models may rely more heavily on private corpora built from a user’s old drafts, notes, emails and tagged social posts. The resulting prose would sound more like that individual. Detection systems could fail not only because models improve, but because the underlying training material is private and inaccessible.
If that happens, the practical question of whether a text was “written by hand” may stop yielding any dependable answer at all.
From literary theory to newsroom reality
The source article ends by returning to Barthes. In 1967, he asked readers to stop centering the author. Nearly 60 years later, AI may force that shift for reasons that are less philosophical than technical. The fight around the Journal, Druckenmiller and AI disclosure suggests that the boundaries around originality, bylines and authorship are already moving.
FAQ
What does Barthes’ “death of the author” mean here?
As presented in the source article, Barthes argued in 1967 that explaining a text through the author’s life and intentions narrows interpretation. He wanted to protect the autonomy of the work and give more interpretive power to the reader.
Does the Wall Street Journal allow outside contributors to use AI in op-eds?
According to Paul Gigot, staff editorial writers still have to write their own pieces. He does not plan to police AI use by outside contributors and does not require a separate disclosure from them.