Real academics, fake citations: AI is now polluting Australia's parliamentary evidence

Divna Haslam, an associate professor and clinical psychologist at the University of Queensland, has spent the past week fielding emails about a study she did not write. The study, supposedly co-authored by her team, was cited as evidence in a submission to an Australian parliamentary inquiry into family violence and suicide. The study does not exist. The submission was filed by Drilldown Reports, a research firm that has since acknowledged using AI in its research process.

The Haslam case is the named-harm anchor of a wider pattern uncovered by Guardian Australia. Australia's parliamentary inquiry process, the channel through which expert witnesses, industry groups, and ordinary Australians feed evidence to the people who write the country's laws, is being flooded with AI-generated submissions that include hallucinated references. Committee reports then cite those submissions, and AI search and chat products (Google AI summary, ChatGPT) treat the hallucinated citations as genuine and pass them back as new evidence, producing a self-reinforcing misinformation loop.

A "hallucination" is the industry term for a confident, plausible-sounding but invented answer produced by a generative AI system. In this case, the invention is a citation that sounds real but does not exist.

Christian Downie, a professor in the Australian National University's school of regulation and global governance, told Guardian Australia the pattern is a "significant risk" to parliamentarians who may end up "making decisions based on evidence that doesn't exist."

Drilldown Reports told the paper it used AI in its research process and that it had identified errors in a follow-up submission, but the deadline to correct the original family-violence submission had already passed. A spokesperson framed the failure as a human upload error while defending AI-assisted research with human quality review. The episode exposes a fixable gap: the inquiry intake desk does not currently verify named-attribution claims against the actual authors, so a fabricated reference under a real academic's name reaches committee members as if it were genuine evidence.

Google AI summary is amplifying the problem. When users search for the topic of the inquiry, the AI summary generates prose that treats the fake paper as if it were a published study. ChatGPT is also citing the inquiry submission, with the fake reference, back to users who ask about the underlying policy area. The result is a closed loop. A hallucinated citation enters parliamentary evidence; the parliamentary submission is then cited as authority by AI products; those AI products create new "evidence" of the hallucinated citation's existence.

This is the third concrete incident in twelve months in which AI-generated material has reached Australian government evidence pipelines. In October 2025, consulting firm Deloitte repaid the Albanese government the A$440,000 fee (roughly US$290,000 at recent exchange rates) for a report that had relied on AI-generated material, including fabricated references. On 17 August 2026, a Senate hearing into Australia's teen social-media ban heard testimony that a report backing the ban contained AI-fabricated citations, as separately reported by Tech Times.

Australia's parliament is in session and the teen social-media ban is under live debate, so the evidence-intake question is policy, not abstract risk. The vulnerability sits at the intake layer of democratic process, not at the AI model layer. Three concrete fixes would close the loop. Parliamentary inquiry secretariats could verify named-attribution claims against the cited author before a submission is entered into the public record. A short email to the named author would catch most Haslam-style errors. Committee reports could require a passed citation check, meaning the cited work exists, the cited author is the actual author, and the cited finding matches the cited text, before a reference is quoted as evidence. AI search and chat vendors that surface inquiry submissions as sources could expose the citation chain, including which submission, which inquiry, and which named author, rather than treating the citation as a black box, so users can verify the chain themselves.

The same gap that put Divna Haslam's name on a study she did not write will keep producing the same result until parliamentary intake verifies the named authors behind the citations it receives.