Updated
Updated · The Guardian · Aug 31
Guardian Finds 39 AI-Tainted Inquiry Submissions in Australia, Citing Fake Research
Updated
Updated · The Guardian · Aug 31

Guardian Finds 39 AI-Tainted Inquiry Submissions in Australia, Citing Fake Research

1 articles · Updated · The Guardian · Aug 31

Summary

  • At least 39 submissions to Australian parliamentary inquiries contained apparent AI-hallucinated references, Guardian Australia found, with some committee reports already citing material whose sources largely do not exist.
  • A Guardian-built checker flagged documents with 20% or more unmatched references against academic databases; more than 100 papers also carried ChatGPT URL tags, suggesting wider AI use than the citation count captures.
  • Real academics including Divna Haslam, Nicole Gurran and Margaret Simons said fabricated citations were convincing enough to misstate or falsely attribute their work, risking policy decisions based on nonexistent evidence.
  • Google AI summaries and ChatGPT can reinforce the problem by summarizing fake references as real and then citing the inquiry submissions themselves, creating a feedback loop of misinformation.
  • The findings revive concerns after Deloitte partially refunded the federal government over fake citations in a A$440,000 report, and experts say new safeguards may be needed without closing the inquiry process.

Insights

When fake AI citations perfectly mimic real academic papers, how can governments prevent fabricated evidence from dictating public policy?
If top consultancies and public submissions are riddled with AI lies, is any government report truly safe from hallucinated data?