NYT reporter used a hallucinated AI quote attributed to Pierre Poilievre

The New York Times needs to explain how a fake quote attributed to Poilievre made it past the editor’s desk and got published.

On April 14, the New York Times’ Canadian bureau chief published a story explaining how Carney won over a majority of voters, and why the opposition — especially Conservative Leader Pierre Poilievre — felt salty about it, despite the whole thing being constitutionally valid.

The problem was not the premise. The problem was the quote.

Pierre Poilievre at the CSFN Photo: Karyn Pugliese

The original copy had Poilievre saying: “If these turncoats have any shred of integrity left, they should resign their seats tonight and run in a by-election tomorrow,” in a March speech. “Let the people decide if they want a Liberal representing them, because they sure didn’t vote for one in the general election.”

Except he didn’t say that. Not like that. Not in those words.

Note that I finally found this in at archive.today It seems the New York Times blocks the Wayback Machine’s IP address so the captures don’t preserve the article — just the access-block screen.

What Poilievre actually said was: “My personal opinion is that when a member of Parliament goes back on the word they made to their constituents and switches parties, constituents should be able to petition to throw them out. That would put the people back in charge of our democracy rather than having dirty backroom Liberal deals by Mark Carney determine our destiny.”

The now corrected quote

The bad quote sat there for weeks. The Times did not correct it until May 1.

The correction raised new questions because it admitted more than an error. The reporters had used AI. And the AI hallucinated a quote Poilievre never said. Somehow, the quote made it into the paper anyway.

The NYT issued this correction on May 1, 2026.

Michelle Cyca — a journalist whose work has appeared in The Narwhal and The Walrus, and a visiting professor at UBC’s School of Journalism, Writing and Media — noted on her Bluesky thread about the incident are that one of the highest-paid reporters in Canada apparently violated the NYT’s own journalism policy on AI, and has survived an error of judgment that would land most Journalism students in the Office of Academic Integrity.

I actually don’t know what an NYT bureau chief makes, but they did just post a Western Canada reporter role with a salary range of $158,000— $235,000 CAD. Most Canadian reporters earn between $52,000 to $62,000 per year, according to the Canadian Journalism Fund, although seasoned or specialized reporters can hit more than $100,000.

But the more important point Cyca made was about a reporter using AI without bothering to fact-check it.

Anyone who has used AI knows what it does. It invents links, sources, quotes, even whole theories that sound real. Then, when pressed, it swears on its little algorithmic heart that the lie is true.

No one should dump data, speeches, reporting notes or other material into AI and trust it to replace their own thinking. Especially not a highly paid reporter.

And as Cyca pointed out, the Times’ own explanatory article, “How The New York Times Uses A.I. for Journalism,” says in the subhead: “We don’t use A.I. to write articles, and journalists are ultimately responsible for everything that we publish.”

It goes on to explain how the NYT does use AI: automated voice translation, scanning large amounts of video, or scanning satellite imagery. Then humans review the work.

That is the point. The Times’ policy emphasizes human review.

I am not against AI. I use it. But AI should be treated like a sloppy assistant you cannot quite bring yourself to fire because it saves time, even though it inevitably fucks up the assignment.

There’s a story I’ve heard about an AI system trained to identify cancer. Programmers fed it hundreds of X-rays from patients with the disease so it could find patterns and learn what cancer looked like.

At first, it seemed to work.

Then it started diagnosing some patients who did not have cancer as if they did. The programmers tried to figure out what the machine was seeing.

It turned out that many of the images where patients were positive for cancer came from the same hospital. Those scans had the hospital’s logo on them.

The AI had not learned to recognize cancer. It had learned to recognize the logo.

Which is why the NYT’s AI policy does not allow AI to be treated as a journalist,a d requires “human oversight and review.”

We are going to see more of this.

Reporters use AI. So do editors. So do you. It is already everywhere.

It takes spam out of our inboxes. It transcribes interviews. It sorts, summarizes, scans and searches.

On days when I have time, I like loading a document into NotebookLM and listening to it as a podcast before I read it. I know it gets things wrong. I know I still have to read the document myself. But a mostly-right audio summary is useful when I am trying to get my head around a dense file.

I am dyslexic, so I use ‘read-aloud tools’ to catch more typos than I would catch on my own. And you should thank me for that use of AI. You would have a hard time reading my copy without them.

I also use AI as a search engine, with a number of rules, including: give me the links to the original documents so I can read them myself.

I am also not above vibe-coding. I never mastered Python, and a gal’s gotta scrape a website sometimes.

We are all trying to learn how to use AI as ethically as possible, without letting it turn us into idiots.

I dread the day I make an error and have to admit AI was involved. It will probably happen, and it will certainly be embarrassing.

But to avoid a career-ending mistake, there are two things I do not do.

First thing: I do not let it write for me.

Actually, no one should.

I kind of hate AI’s prose. The over-explaining. The tidy little opposites. The mixed metaphors and clichés. Those naggy wrap-ups at the end of paragraphs, telling you one more time what you were already smart enough to understand.

Second thing: I do not let AI think for me.

Being someone who uses AI, I am not going to ride in on the high horse, pointing fingers and screaming, “How dare you use AI?”

The issue is that the NYT seems to have a solid policy, but I can't tell from the thin explanation on the correction page what actually happened.

Did the reporter feed AI a transcript and let it pick clips (Why would anyone do that, knowing AI hallucinates? Besides, pulling a quote from a transcript is not exactly heavy lifting.) How did a hallucinated quote make it past the reporter and the editor’s fact check? What changes will be put in place to prevent this from happening again?

Because this is not a small thing.

It was a quote from a national political leader. That is not filler text. Careers change over what people are believed to have said. So can a country’s history.

This hallucinated error was bad. It changed the tone. It made Poilievre sound more volatile in his anger.

It could have been worse.

It could have been more inflammatory, more damaging, or further from what he actually said.

Maybe the next hallucinated quote will be worse.

Maybe it will not be about Canadian floor-crossing.

Maybe it will be about Russia and Ukraine, Israel and Gaza, Lebanon, Iran, or something the U.S. president is believed to have said in the middle of a crisis. We cannot afford invented words in moments like that.

And as public trust in media and institutions keeps falling, transparency is the only way you can stop one mistake from becoming proof of every suspicion and fear people already have about you.

Before AI, fabricating a quote was generally understood to be intentional and grounds for dismissal. With AI, fabrication is unintentional, but it is still a catastrophic mistake.

A mistake that serious should not vanish into a correction note while the byline carries on untouched and unexplained.