Full Breakdown
AI-Generated Hallucinations Undermine Australian Parliamentary Inquiries
9/1/2026, 8:10:41 AM
The Core Issue
Government-commissioned inquiries rely on written submissions to inform policy. Recent analysis reveals that dozens of these submissions contain fabricated academic references and other misinformation produced by large-language models (LLMs) such as ChatGPT and Claude. The false citations appear as plausible studies, leading committees to consider evidence that does not exist.
Detection Methodology
The Guardian built a custom program that extracted every reference from inquiry documents submitted to the current parliament and cross-checked them against CrossRef, Google Scholar and DOI resolvers. Submissions in which 20 % or more of the references could not be matched were flagged for manual review. The investigation also scanned links for metadata tags automatically added by ChatGPT.
Scale of the Problem
- At least 39 submissions were found to contain hallucinated references, a conservative estimate because the method only identifies documents with citation errors.
- More than 100 papers included ChatGPT-generated URL tags, indicating the platform’s involvement in the reference lists.
- Submissions ranged from a few incorrect citations to documents where every reference was fictitious.
The submissions originated from individuals and organisations across the political spectrum, showing that the issue is not confined to any single party or ideology.
Official and Expert Responses
He emphasised the need for new guidelines to “encourage truthfulness” in the inquiry process.
“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.” — Divna Haslam
Industry reaction – Drilldown Reports, the firm that authored one flagged submission, acknowledged using AI in its research workflow and said it had identified the errors in a follow-up document, but the original deadline had already passed.
Academic concerns – Divna Haslam, associate professor of clinical psychology at the University of Queensland, described a fabricated reference attributed to her team as “scary” because it could mislead a cursory reader. She stressed that “we just can’t risk that” in the domestic-violence policy arena.
Potential Implications
- Policy risk: Mis-attributed research could shape legislation on sensitive issues such as family violence, climate change and housing, where evidence-based decisions are critical.
- Public trust: Repeated exposure to fabricated citations may erode confidence in parliamentary processes and the institutions that underpin democratic decision-making.
- Regulatory gap: Current Senate advice warns that submitters remain responsible for accuracy, but no systematic detection tools are mandated for all submissions.
Why It Matters
The infiltration of AI-generated hallucinations highlights a broader challenge for democratic institutions: ensuring that rapidly accessible generative technologies do not compromise the evidentiary standards required for sound governance. As AI tools become more embedded in research and drafting workflows, the need for verification protocols and clearer accountability mechanisms grows.
Verbatim Quotes
- “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.” — Divna Haslam
