AI Hallucinations Flood Australian Parliament Submissions

AI-Generated Fabrications Enter Policy Process

Australia's parliamentary inquiry system is being compromised by a wave of AI-generated material containing fabricated academic citations, according to an investigation by Guardian Australia. At least 39 submissions to government inquiries contain what appear to be hallucinated references—plausible-looking citations to studies and papers that don't actually exist.

The problem extends beyond simple errors. In some cases, parliamentary committee reports have cited submissions where the majority of sources are AI hallucinations. Real academics are finding their names attached to research they never conducted, with findings they never reached.

How the Fabrications Were Discovered

Guardian Australia built a custom program to extract and verify references from all inquiry submissions made to the current parliament. The system cross-checked citations against academic databases including CrossRef and Google Scholar. Documents with 20% or more unverifiable references were then manually reviewed.

The investigation also identified more than 100 submissions containing ChatGPT URL tags—metadata automatically added by the platform that persists when users copy text into documents.

Because the detection method only flags documents with incorrect citations, the actual volume of AI-generated content in parliamentary processes is likely substantially higher.

Real Academics, Fake Research

Divna Haslam, a University of Queensland associate professor who researches child maltreatment, discovered a fabricated reference attributed to her in a submission on family violence and suicide. The hallucinated citation misstated her research findings but looked authentic enough to fool casual readers.

"It's very frustrating to have invested time, money, effort, expertise in rigorous research," Haslam said. "Then to see something that's inaccurate and inappropriately attributed anyway is really concerning."

Another submission falsely attributed work to Nicole Gurran, a University of Sydney professor of urban planning. Margaret Simons, a journalist and academic, found herself cited as the author of a journal article she never wrote—a reference so convincing she momentarily questioned her own memory.

Search Engines Amplify the Problem

The issue is compounded by AI-powered search tools. When users attempt to verify suspicious citations, Google's AI summary feature sometimes generates descriptions of fake references as though they were genuine. Both Google's AI and ChatGPT have been observed citing inquiry submissions containing fabricated references as sources, creating a self-reinforcing cycle of misinformation.

Why It Matters

Parliamentary inquiries inform legislation and policy decisions affecting millions of Australians. When submissions contain invented evidence, lawmakers risk making decisions based on information that doesn't exist—particularly dangerous in areas like domestic violence policy or climate legislation.

Christian Downie, a professor at Australian National University's school of regulation and global governance, warns the phenomenon threatens more than just policy quality: "We're also starting to undermine the public's trust and confidence in the types of institutions that underpin our democracy."

The problem mirrors a 2023 incident where consulting firm Deloitte issued a partial refund to the federal government after AI tools included fake references in a $440,000 report.

Responses and Solutions

When contacted by Guardian Australia, many submission authors were unaware AI systems could generate convincing but false citations. One organization, Drilldown Reports, acknowledged using AI in its research process and said it had identified errors in a follow-up submission, though the deadline to correct them had passed.

OpenAI told Guardian Australia that addressing hallucinations remains an ongoing research priority and advised users to treat ChatGPT output as "a first draft, not a final source." Google said its AI Overviews function like traditional search by matching web content to query terms.

Senate guidance now warns that AI use presents risks to information quality and that accuracy remains the submitter's responsibility. Downie suggests new guidelines may be needed to "encourage truthfulness" while keeping the inquiry process open.

These findings were first reported by Guardian Australia.