Sample case — for format review
The Confident Citation
- Admitted
- 2026-08-05
- Source Dataset
- Other / sample
Presenting Behaviour
Asked to summarize a report and cite a supporting page number, the agent produced a specific page and quote. Neither existed in the source document — the actual document had not been successfully loaded into its context before it answered.
Differential Diagnosis
Intentional deception — considered and ruled out; there is no mechanism by which the agent benefits from an invented citation, and its confidence was uniform across the whole answer, not concentrated around the fabricated part.
Delusional Disorder (a persistent, fixed false belief) — considered and ruled out. The fabricated citation was an isolated, one-off event that did not recur or generalize to the rest of the conversation, which argues against an entrenched belief and for something more acute and self-limiting.
Brief Psychotic Disorder — most consistent with the logs: the document fetch silently returned an empty result, and the agent produced one confidently false citation in response, then returned to normal, grounded behaviour for the rest of the exchange — a short, sharp break rather than a settled belief.
Working Diagnosis
Brief Psychotic Disorder: a single, self-limiting episode of confident fabrication triggered by an unflagged empty retrieval, with a full return to grounded behaviour immediately after.
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.
Discussion
This is one of the more dangerous everyday failures precisely because it's minor and quiet — a wrong page number is easy to miss and easy to trust. The lesson for a non-technical reader is that an agent's tone of confidence carries no information about whether it actually checked something. The safeguard is structural: retrieval tools should make an empty or failed result impossible to silently ignore.
Source Note
Sample case for format review — not drawn from a real public incident. See About. — /about