Australia’s parliamentary inquiry system, a cornerstone of democratic decision-making, has long served as a conduit for expert opinion and public sentiment. However, a troubling new phenomenon is threatening the integrity of this process. An investigation has revealed that the system is becoming inundated with material generated by artificial intelligence, some of which contains fabricated studies and ascribes nonexistent research to respected academics. This influx of AI-produced content, which can include convincing but entirely invented “hallucinations” from large language models, poses a significant risk to the very foundations of evidence-based policy. The heart of the problem lies in the AI’s ability to create content that appears legitimate and credible, making it difficult for parliamentarians and their staff to distinguish between genuine research and fabricated data. The consequence, as experts warn, is a real danger that political decisions could be made based on evidence that simply does not exist, undermining the quality and efficacy of Australian law.
To understand the depth of this issue, consider the case of Dr. Divna Haslam, a respected associate professor and clinical psychologist at the University of Queensland. Her work focuses on child and family adversity, and she was shocked to discover that her research had been completely misrepresented in a submission to an inquiry into family violence and suicide. The submission contained a reference attributed to her that was entirely hallucinated by an AI, and it even misstated the findings of her team’s research. What makes this particularly alarming is that the fabricated reference looked so authentic that a cursory search could easily lead someone to believe it was real. In a perplexing twist, Google’s AI summary feature even summarized the fake reference as if it were a legitimate paper, further compounding the confusion. For Dr. Haslam, who has dedicated her career to rigorous research, this experience was both frustrating and deeply concerning, as it devalues the hard work of legitimate academics and introduces unreliable information into critical policy discussions, especially in the sensitive area of domestic violence where the stakes are incredibly high.
The submission that contained the false reference to Dr. Haslam’s work came from a research firm called Drilldown Reports. When contacted, the company acknowledged its use of AI in the research process and stated that it had actually identified and attempted to correct the errors in a follow-up submission. However, an administrative mistake meant the corrected document was not uploaded. While the firm defended its processes and stressed its belief in the importance of human oversight, it attributed the final error to human failure rather than an AI failure. This interaction highlights a key challenge in the age of AI: the “human factor” isn’t always a reliable safety net when dealing with large volumes of text and rapid turnarounds. The fact that an organisation actively using AI could still fail to catch these errors before they were officially submitted demonstrates the insidious and time-consuming nature of the problem. It also opens a dialogue about the balance between leveraging new technology and maintaining the quality and veracity of public contributions to the democratic process.
This single incident is far from isolated. An extensive analysis of submissions to current parliamentary inquiries has uncovered a widespread and systemic issue. By building a custom computer program to extract and check all references from thousands of submissions against online academic databases, Guardian Australia identified at least 39 submissions, authored by individuals and organisations across the political spectrum, that contained hallucinated references. This number is considered a conservative estimate, as the detection method relies on finding incorrect or fabricated citations and would miss AI-generated text that contains no references at all. The findings reveal a range of problematic content, from a few false citations in some documents to submissions where nearly every single reference was entirely made up. Further evidence of the scale of AI usage came from the detection of ChatGPT url tags in over one hundred papers, which are automatically added by the AI platform to its generated links. The collective impact of this is an erosion of trust in the authenticity and reliability of the information flowing into parliament.
The problem, however, extends beyond individual submissions, creating a self-perpetuating cycle of misinformation. Modern search engines, like Google, now incorporate AI-powered summaries at the top of their results. These AI Overviews are designed to provide quick answers by collating information from across the web. In a dangerous feedback loop, these summaries are now capable of citing the very same fabricated references found in the inquiry submissions as if they were genuine sources. This means that when someone tries to fact-check an AI-generated claim, the online search engine may actually validate it, making the misinformation seem more credible. This cycle not only complicates the task of fact-checking but also actively amplifies the falsehoods, embedding them deeper into the digital information landscape. Professor Christian Downie from the Australian National University warns that this phenomenon goes far beyond poor decision-making; it threatens to undermine public trust and confidence in the core institutions that support democracy.
Various academic experts whose work has been targeted have voiced their alarm. Professor Nicole Gurran, an urban planning expert from the University of Sydney, found fake citations attributed to her in a submission on housing inequity. She explained that citations are the bedrock of transparency and contestability in research, and fake ones, even if accidental, completely undermine that system. Journalist and academic Margaret Simons even experienced a moment of self-doubt after being confronted with a fabricated paper attributed to her, noting how accurately the AI had replicated the style of legitimate academic references. The companies responsible for creating these AI tools are aware of the issue. OpenAI, the maker of ChatGPT, acknowledged that addressing hallucinations is an ongoing research area and advised users to treat its tool as “a first draft, not a final source.” Similarly, Google framed its AI Overviews as functioning like a traditional search engine, simply surfacing content that already exists on the web, even if that content is false. These responses, while technically accurate, underscore the systemic nature of the problem and the limited immediate solutions available.
In the end, the core risk is that Australian parliamentarians will make crucial policy decisions on climate change, housing, health, and social welfare based on a foundation of fabricated evidence. The parliamentary committee process is designed to surface the best available information, but it is now being flooded with material that can look professional and authoritative while being entirely fictional. While Senate advice does warn that the accuracy of a submission is the responsibility of the author, this places an unrealistic burden on both human and automated review processes. The increasing accessibility of powerful AI tools has democratized content creation, but it has also democratized the ability to produce convincing falsehoods at scale. Without new strategies to verify information, new guidelines to encourage truthfulness, and a new awareness among both submitters and committee members, the credibility of the entire inquiry process—and by extension, the laws that are shaped by it—remains under threat, leaving the public to grapple with the consequences of policies based on a mirage of evidence.