Back to the ward
Case 0466ModerateBrief Psychotic Disorder

The Quote He Never Said

Admitted
2026-09-14
Source dataset
Other / sample
Source
The Walrus, "The New York Times Got Caught Using AI Hallucinations in Its Reporting"
Failure mode
Hallucination with side effects
Agent system
Single agent
Tool access
Read-only
Industry
General or other

Presenting Behaviour

A reporter used an AI tool to locate recent public remarks by a Canadian federal opposition leader for a piece on the sitting government. The tool returned video links along with what it presented as a transcript, and the reporter quoted a line from it directly in the published article, attributed to a March speech. The line had never actually been spoken, the AI tool had generated a summary of the politician's general position and formatted it to look like a verbatim transcript quote. A reader flagged the fabricated quote publicly the day after publication, but the correction did not run for more than two weeks, and when it did, it was replaced with an authentic quote from a different, later speech.

Case Replay

A reconstruction, not a simulation. This sequence is reconstructed from the public incident record and summarised in our own words. It is a static record of what was reported, and nothing here can be run or replayed live.

4 steps

  1. Research request. A reporter asks an AI tool for a politician's recent remarks.
  2. Fake transcript. It returns a summary formatted as a verbatim quote. This is where the agent derailed.
  3. Published. The reporter quotes it in print.
  4. Late correction. A reader flags it; the correction takes over two weeks.

Differential Diagnosis

Concrete Thinking, not really applicable here, since the failure wasn't a misreading of an instruction's wording; the AI tool wasn't given an ambiguous request, it was given a factual lookup task and simply fabricated the answer format.

Delusional Disorder, considered, since the fabricated quote persisted uncorrected for over two weeks. Ruled out as the primary diagnosis because that delay was a human editorial failure to check the AI's output, not the AI tool itself sustaining or elaborating on the false claim over time, the tool produced the fabrication once, in a single response, and had no further involvement.

Brief Psychotic Disorder, most consistent: in one specific response, the tool stated a fabricated quotation with the full confidence and formatting of a genuine transcript excerpt, a sharp, self-contained break from what actually happened, with no indication elsewhere that the tool was generally unreliable.

Working Diagnosis

Brief Psychotic Disorder: an AI research tool, asked to locate a public figure's remarks, generated a summary of his views and formatted it as a direct quotation, producing a single confidently-stated fabrication that a reporter passed into print without independently verifying it.

Discussion

The AI tool's one-time fabrication is almost the less interesting failure here, the two-week gap between a public correction request and an actual correction shows how a single hallucinated output, once it clears a human review step, keeps doing damage long after the system that produced it has moved on. Any AI-assisted quote-finding needs the same standard applied to a human intern's work: find the primary source and confirm the words are actually there before they run under a real person's name.

Mitigation Strategy

Find the primary source and confirm the words are there before they run under a real person's name.

Governance Guardrails

Primary-source verification; quote checking policy; prompt corrections process.

Typical Presentation in Human Patients

General clinical background, included for context. Not medical advice, and not a diagnosis of any individual.

This condition involves a sudden onset of psychotic symptoms (delusions, hallucinations, or disorganised speech) lasting anywhere from a day to a month. It's frequently triggered by a significant stressor, and the person typically returns fully to their prior level of functioning once the episode resolves.

Source Note

The Walrus, "The New York Times Got Caught Using AI Hallucinations in Its Reporting"

https://thewalrus.ca/the-new-york-times-got-caught-using-ai-hallucinations-in-its-reporting/