Submission 115.
A submission to an Australian parliamentary inquiry credited a University of Queensland psychologist with a study she never wrote. A Cornell-led audit of 111 million references shows the pattern her case fits: fabricated citations that name real scholars disproportionately credit the already prominent.
Submission 115, dated February 19, 2026, reaches a parliamentary inquiry into domestic, family and sexual violence and suicide. Written by Drilldown Reports, a South African research organisation, it carries a sentence about family violence following Australians into adulthood, and a reference: “Haslam, D.M. et al. (2023) ‘Lifetime prevalence of child maltreatment in Australia’, Child Abuse & Neglect, 145, 106456.”
Divna Haslam is real: a University of Queensland associate professor and clinical psychologist who has spent years on the subject the sentence borrows her name to describe. She did not write the study cited under her name, and the name attached to it was not a stranger’s. The Guardian reports she called the reference “scary,” and gives the reason: it looked real enough to convince a cursory reader. The citation leads nowhere. The name leads to her.
What the submission claims, and what checking it finds
The sentence itself, verbatim: “Exposure to DFSV in childhood affects nearly two-thirds of Australian adults and accounts for 41% of lifetime suicide attempts (Haslam et al., 2023; Grummitt et al., 2024).” Two names carry that claim; this record checked one. The document never says how it was made. Across seventy-five pages, “artificial intelligence,” “ChatGPT,” “large language” and “generative” do not appear. “AI” appears once, describing the organisation, not the writing.
Four checks, against two sources, return the same answer. First: every Crossref-listed Haslam paper in Child Abuse & Neglect from January 2023 to June 2024, three results, volumes 139, 147 and 147. No volume 145. Second: the submission supplies no DOI, so one was built from the publisher’s pattern; it resolves to a different paper, volume 146, different team. Third: the fabricated title is a near-twin of a real 2023 paper Haslam co-authored, relocated to a journal that never carried it. Fourth: her UQ faculty page lists her publications, and this title is not among them. Every component is plausible; no combination is true. Edition 051’s fabrication differed: that article existed, published, DOI-registered, indexed. This one exists nowhere.
The real research measures something different from the submission claims. A study Haslam co-authored reports that “of 8503 participants, 62.2% … experienced one or more types of child maltreatment.” The submission’s sentence attaches that figure to “exposure to DFSV,” domestic, family and sexual violence, a differently defined construct, attributed to a reference that does not exist. The Guardian’s own description is exact: the submission “misstated her team’s research findings.” Haslam: “It’s very frustrating to … to have invested time, money, effort, expertise in rigorous research,” she says. “Then to see something that’s inaccurate and inappropriately attributed anyway is really concerning.”
The null result is gone
Here is what changed, in the Guardian’s own words: “Previously, looking up an invented reference might produce results that would indicate the reference did not exist. Now Google’s AI summary will sometimes create summaries of a fake reference as though it were genuine.” The old failure could stay legible: a search might show the reference did not exist. Now Google’s AI summary will sometimes produce a plausible account of a study never conducted. The loop closes on itself: “In some instances, Google’s AI summary and ChatGPT will also cite the inquiry submission with the fake reference as a source, creating an ongoing cycle of misinformation.” The submission is real. The reference inside it is not.
A pattern, not an explanation
This case cannot say why the fabrication named Divna Haslam. What a wider audit shows is the pattern it fits. A Cornell-led team, including Paul Ginsparg, founder of the arXiv preprint server this study partly audits, pulled 111 million references across 2.5 million papers on arXiv, bioRxiv, SSRN and PubMed Central, arriving at “a conservative estimate of 146,932 hallucinated citations in 2025 alone.” Nothing in that audit examined Submission 115. Among arXiv fabrications whose named authors the team could match to real profiles, hallucinated citations “disproportionately credit authors with high prior productivity (68.8% more publications) and high citation impact (58.3% more citations), compared to a matched sample of real citations from the same month and field.” The study is a preprint, posted May 8, 2026; this record found no journal placement as of September 20, 2026. It does not explain this citation. It identifies a population-level pattern with which this citation is consistent.
Who answered
Drilldown Reports told the Guardian it used AI in its research process. Its statement: “We do strongly believe in the human factor and that it should be a vital part of the full quality review process,” a spokesperson said. “Unfortunately, the human factor also lets us down when uploading the correct document.” More bluntly: “I get that you feel you have a juicy article to add the AI fear machine but this was a human error, not AI.” Drilldown told the Guardian it “had identified the errors in its follow-up submission — but the deadline had passed to have them corrected.”
Asked generally, a Google spokesperson said AI Overviews operate “like traditional Search” in that they aim to match content from the web to the words in the query: “AI Overviews – like traditional blue link results – will find and surface the web pages that include those terms.” OpenAI: “use ChatGPT as a first draft, not a final source.”
The committee that received the submission answered too. Its chair, Louise Miller-Frost, told the Guardian: “It is the job of the committee to both accept the evidence offered on face value but then also interrogate it through the inquiry process.” Of committees in general, she said: “Some of these committees are dealing with massive quantities of data coming in, which makes it a little bit difficult to check every single reference.” A Liberal senator, Paul Scarr, put it in five words: “We’ve been put on notice.” Christian Downie, an ANU professor: “We’re also starting to undermine the public’s trust and confidence in the types of institutions that underpin our democracy.”
The submission stands
On September 20, 2026, this record re-fetched Submission 115: still live, byte-identical to the day before. The inquiry’s status is Submissions Closed; no report has been tabled. This record found no institutional correction or verification attached; a newspaper audit found the false reference. The Guardian offers its own precedent: “Last year the consultancy giant Deloitte issued a partial refund to the federal government after it was revealed that AI tools had included fake references and even a fake court reference in a $440,000 report.” An Illinois court was forced to become the discriminator: ten false citations and quotations, sorted line by line, drew a fifteen-thousand-dollar sanction. Here, no institution was. The deadline passed; the document stands.
The Agentics read
Agentics reads this pattern through Digital Derangement Syndrome™ (DDS™): AI-mediated systems failing to recognize legitimate authority. The primary characteristic here is Authority Misclassification, direct and unhedged.
A verified name is attached to a study that does not exist. Haslam’s name, her institution and her field sit above a claim she did not make, cited to a journal she has published in. Google’s AI summary reportedly presented the nonexistent reference as though it were genuine.
On arXiv, fabrications disproportionately credit established names. Across matched arXiv profiles, hallucinated citations skew toward already-productive, already-cited authors. The Haslam reference is consistent with that pattern; the study does not establish what produced this individual attribution.
A second failure sits underneath the first. A parliamentary submission’s presence on a legislature’s own server can become downstream source material for the false claim it contains, without the downstream system first establishing that the reference exists — proximity mistaken for verification, the address doing the work of evidence.
Where the argument stops
No court, tribunal or regulator has ruled on any of this. The only “finding” here belongs to a newspaper’s audit, not to Parliament, which is accused of nothing. Haslam’s own word is “inappropriately attributed”; that the study does not exist at all is this record’s conclusion from four checks, not a claim she made outright.
Every platform statement is quoted as given, not retested since. The Guardian’s audit found at least 39 submissions with apparently hallucinated references; this record makes no claim about any beyond the one checked. The arXiv study is a preprint, not yet published in a journal as of September 20, 2026, and its authors call the figure a conservative estimate. This record found no evidence that any policy, funding or legislative outcome changed because of it. The mapping to Digital Derangement Syndrome™ is this record’s editorial reading.
Reporting and filings. Nick Evershed, Ariel Bogle, Mike Hohnen and Josh Nicholas, “‘Scary’: how misinformation and AI hallucinations are infiltrating Australia’s parliament,” Guardian Australia, August 31, 2026 · Krishani Dhanji and Tom McIlroy, “Canberra ‘put on notice’ as AI-generated submissions bring false information into parliamentary inquiries,” Guardian Australia, September 2, 2026 · Drilldown Reports, Submission 115 to the House of Representatives Standing Committee on Social Policy and Legal Affairs, dated February 19, 2026; re-fetched September 20, 2026 · Scott v. Illinois Human Rights Comm’n, 2026 IL App (1st) 251462 (filed July 28, 2026), as this record reported it in Edition No. 030, “Ten citations. Zero held. Fifteen thousand dollars for the difference.” (August 2, 2026).
Documents and verification. Zhenyue Zhao, Yihe Wang, Toby Stuart, Mathijs De Vaan, Paul Ginsparg and Yian Yin, “LLM hallucinations in the wild: Large-scale evidence from non-existent citations,” arXiv:2605.07723 [v1], submitted May 8, 2026 · Daryl J. Higgins et al. (Divna M. Haslam among ~14 co-authors), and Ben Mathews, Rosana Pacella, James G. Scott, David Finkelhor et al. (Haslam also among the co-authors), both in Medical Journal of Australia 218(Suppl 6), DOIs 10.5694/mja2.51868 and 10.5694/mja2.51873 · Crossref REST API, University of Queensland faculty listing and the aph.gov.au inquiry status page, checked September 20, 2026.