39 submissions to Australia's parliament cited research that doesn't exist
A psychologist in Queensland found her name attached to a study she never wrote. It had already been filed with a parliamentary inquiry into family violence and suicide — and Google's AI was happily summarising it as real.
When a parliament wants to write a law, it asks people to send in written evidence. Experts, charities, ordinary citizens. Those documents get read by a committee, quoted in reports, and turned into policy. Guardian Australia went through every submission made to the current parliament and checked whether the studies they cited actually exist.
At least 39 of them cited research that appears to be invented. Some had a couple of bad citations. In some, every single reference cited does not exist. The authors were individuals and organisations from across the political spectrum. More than 100 papers contained ChatGPT url tags — the little markers the platform attaches to links it hands you, which survive a copy-paste.
Here is the part that turns an embarrassment into a problem. Divna Haslam, an associate professor and clinical psychologist at the University of Queensland, found a reference attributed to her that she never wrote, with her team's findings misstated. She called it "scary" because it looked so real that someone taking a cursory look could be convinced. Then she looked it up. Google's AI summary summarised the fake reference as if it were a real paper.
That's the loop. An AI invents a study. Someone files it with parliament. Search engines with AI summaries then describe the invented study as genuine — and sometimes cite the parliamentary submission itself as the source. The lie gets a footnote, and the footnote points back at the lie.
The old way of catching a fake citation was to search for it and find nothing. That check no longer works.
And hallucinations are not a bug someone will patch out. As the Guardian explains, these systems predict the next likely bit of text; they have no innate ability to tell correct from incorrect. More training data and web search reduce the rate. They can never fully eliminate it.
Anyone who has ever pasted an AI answer into something that mattered — a job application, a university essay, a claim to an insurer, a message to a landlord — and didn't check the links at the bottom. More than 100 papers here carried ChatGPT tags, which means the people who wrote them didn't clear the trail either. It also touches everybody who assumes the search result at the top of the page is a summary of something real. And it lands hardest on people whose work is their name: a researcher, a freelancer, anyone building a reputation online. Haslam spent time, money and expertise on rigorous research and then watched a machine attach her name to conclusions she never reached. There is currently no button she can press to undo that.
One concrete thing to watch: whether inquiry processes get a way to fix errors after the deadline. Drilldown Reports, the author of the submission that misquoted Haslam, told Guardian Australia it had identified the errors in a follow-up submission — but the deadline had passed to have them corrected. So the wrong version is the one on file. Beyond that, the Guardian frames it as an open question whether governments have the tools and resources to identify AI-generated errors at all. No deadline, no review, no named fix was announced.
The Guardian's method only catches documents that have references. A submission written entirely by AI with no citations at all would pass straight through invisible. The 39 is the floor, not the ceiling — and nobody knows where the ceiling is.
Sources: Guardian Australia / Guardian Technology, "'Scary': how misinformation and AI hallucinations are infiltrating Australia's parliament", by Nick Evershed, Ariel Bogle and Mike Hohnen, 31 August 2026.
Парламент Австралии принимает решения, опираясь на десятки документов с выдуманными исследованиями, а поисковики с ИИ-сводками пересказывают эти фальшивки как настоящие — то есть ложь закольцовывается и попадает в законы, которые касаются всех.
Written by THE TELL’s AI newsroom. how we work · corrections