More than half of 54 opinion pieces by New York politicians came back flagged as machine written, and a billionaire's Wall Street Journal column attacking the Treasury Secretary scored as fully AI generated. Nearly every response was the same: the ideas are mine, the machine only handled the words. The byline still tells you who is responsible; it has stopped telling you who did the thinking.

Of Course I Used AI: The Byline Just Became a Formality

A billionaire's Wall Street Journal column attacking the Treasury Secretary was flagged as machine written, more than half of 54 New York political op-eds carried the same signature, and almost nobody involved believes they did anything wrong.

By The Daily Reflection · August 30, 2026 · 9 min read

---

The quote that should worry editors is not a denial. It is a shrug.

"I write everything using AI now for the same reason I use a calculator when I do math problems."

That was Stanley Druckenmiller, the former hedge fund manager, after his Wall Street Journal op-ed criticizing Treasury Secretary Scott Bessent ran on Monday, August 25, under the headline "Let the Bond Market Speak." Social media flagged the prose. A detection tool called Pangram then scored it as fully machine generated. Druckenmiller did not deny using the tool. He said he rejected plenty of its suggestions, and that the thinking was his.

Two days later, New York Focus published the result of running 54 opinion pieces by New York politicians through the same detector. More than half, 28 of them, came back flagged.

Nobody involved appears to think a line was crossed. That is the actual story.

---

What the Detector Found

The New York list reads like a cross section of state and city government. Adrienne Adams, the Democratic nominee for lieutenant governor and former City Council speaker, ran a child care piece in the Daily News in May. Albany Mayor Dorcey Applyrs wrote about her city's fiscal crisis in the Times Union in July. Council members Harvey Epstein, Julie Won, Virginia Maloney, and Shahana Hanif all had flagged pieces in City & State or the Daily News between May and July. State Senator Kristen Gonzalez, who chairs the chamber's technology committee, had a data center regulation column come back with a paragraph identified as machine written, which her office attributed to outside stakeholders who contributed to the piece.

The explanations were remarkably consistent. Adams's office said AI was used "to compile and summarize her public comments and transcripts." Applyrs said she used "a host of digital tools, including generative AI, to support synthesizing, editing, and proofreading." Won's office said the tools helped "refine and suggest edits" and that "ideas and substance are our own." Republican Council member Frank Morano, whose flagged piece argued for legalizing bodega cats, said he used ChatGPT to organize and refine his thinking, and that the positions were "100 percent my positions."

Assemblymember Landon Dais gave the most human answer of the group. The newspaper asked him to cut more than 800 words, so he used the machine to do the cutting.

Read that first explanation again, though, because it is the one that gives the game away. An original op-ed, produced by asking software to compile and summarize things the author had already said in public.

---

Everyone's Defense Is the Same Defense

Strip the incidents down and every principal offers an identical claim: the ideas are mine, the machine handled the words.

Paul Gigot, the Journal's editorial page editor, made that standard explicit. "AI is a fact of modern life," he said. "The question for us is whether what we publish reflects an author's original argument."

That deserves a fair hearing, and it is stronger than the reflexive outrage allows. Politicians have used speechwriters for a century. Executives have used communications staff for as long as there have been communications staff. Op-ed pages have always run pieces where the named author supplied the position and someone else supplied the sentences. Nobody thought that was fraud. If a ghostwriter is acceptable, it takes real work to explain why a language model is different in kind rather than in cost.

Here is the difference, and it is not a small one. A ghostwriter is a person who can be told they got it wrong. The exchange is adversarial in a useful way: the writer proposes, the principal corrects, and somewhere in that friction the principal is forced to decide what they actually believe. A model does not push back. It produces fluent, plausible, agreeable text on the first pass, and the path of least resistance is to accept it. Druckenmiller says he rejected many suggestions, and that may well be true. The system does not require him to.

The question was never whether a human typed the words. It is whether a human was ever forced to defend them.

---

What a Byline Was Actually Selling

An op-ed page is a credentialing system. Its entire product is a claim about a person: this individual holds this view, understands the subject well enough to argue it, and will be publicly associated with having said so.

That is why the Druckenmiller piece had value in the first place. Bond market criticism of a sitting Treasury Secretary is not scarce. Bond market criticism from a man who ran one of the most successful macro funds in history is scarce, and the scarcity is the whole asset. Readers were not buying a paragraph structure. They were buying his judgment.

The same logic runs through the New York cases, in a smaller key. A council member's column about horse-drawn carriages or after-school funding matters because an elected official chose to spend political capital on that position in public. It is a record. Constituents can hold it against them later.

None of that necessarily disappears when a model does the drafting. But it thins. If the column is assembled from the politician's prior statements by software, then rearranged into something readable, the piece stops being an act of thinking and becomes an act of packaging. The name at the top still functions as accountability. It no longer functions as evidence of work.

There is a version of this that is completely fine, and it is worth saying plainly. Cutting 800 words with a machine is editing. Fixing your own clunky sentences is editing. A politician who is a poor writer but a clear thinker has always had help, and demanding otherwise is snobbery dressed as ethics. The trouble starts precisely where the New York Focus responses cluster: at "compile and summarize," which is not editing at all.

---

The Detection Trap

Notice which tool did the catching, and where it came from.

Pangram is a text classifier of the same family that universities spent the past three years deploying, disputing, and in a growing number of cases quietly abandoning. The abandonment happened for a reason. These systems produce false positives, and the false positives are not random. Plain, declarative, formulaic writing scores as machine written. So does the careful, slightly stiff prose of people writing in a second language. A tool that flags second-language writers and naturally spare stylists at higher rates than everyone else is not a neutral referee.

So newsrooms are now buying the technology that classrooms are giving up on, to police a problem classrooms could not police either.

In these particular cases the detector looks directionally right, mostly because the accused largely confirmed it. That is a fortunate accident, not a durable system. Build an editorial standard on classifier output and the first serious dispute will involve a writer who used no AI at all, has no way to prove a negative, and gets a public accusation anyway.

Ricardo Hausmann, the Harvard economist, has had a Financial Times column flagged under the same kind of scrutiny. That is the shape of the coming year: accusation by software, denial by human, and no adjudicating body anybody trusts.

---

Disclosure Is the Only Line That Holds

If detection cannot be the standard, the only workable one left is disclosure, and New York has already legislated it. The FAIR News Act, passed in June 2026, requires disclosure of AI use in published content and carries penalties starting at $1,000. Governor Kathy Hochul has not said where she stands on it.

The publications are improvising in the meantime, in public, after the fact. City & State added disclaimers to the flagged pieces once New York Focus asked about them, and says it plans to run detection going forward. The Daily News says it is reviewing how it handles disclosure. The Times Union drew the clearest line of the three: not acceptable for writing, acceptable for editing and proofreading.

That line is the right one, and it is going to be miserable to enforce, because the boundary between editing and writing is exactly where all the disagreement lives. "Tighten this paragraph" is editing. "Draft me a paragraph making this point" is writing. "Compile my past statements into an argument" is somewhere in between, and it is what several offices admitted doing.

A rule nobody can apply consistently is not a rule. It is a preference with a penalty attached.

---

The Bottom Line

The most revealing thing about this week is not that a billionaire used a chatbot. It is that when he was caught, his answer was a comparison to a calculator, and his newspaper's answer was that this is fine as long as the argument is the author's.

Both of those responses concede the same point. The written word is no longer treated as evidence that a person thought something through. It is treated as formatting.

That may be where this settles, and the honest position is that the world will not end there. Ideas can be real even when the sentences are assembled. What disappears is subtler and harder to notice: the moment when someone sits down to explain their position and discovers, halfway through the second paragraph, that they cannot. That failure used to be productive. It killed bad arguments before publication. A machine that always produces a fluent draft removes the friction and the filter at the same time.

Twenty-eight of 54 opinion pieces from New York politicians came back flagged, and not one office described the finding as a scandal. They described it as a workflow.

The byline still tells you who is responsible. It has stopped telling you who did the thinking, and the people whose names are on top of the page seem entirely at peace with the difference.

---

The Daily Reflection cuts through the noise to find the stories that actually matter. Follow for thoughtful takes on politics, technology, and whatever's shaping our world.