AI Hallucinations Infiltrate Australian Parliament Inquiry Submissions

AI-Generated Content in Parliamentary Submissions

Guardian Australia revealed that Australia's government inquiry process is being flooded with AI-generated material. These submissions invent studies and attribute nonexistent research to real academics and authors, impacting the integrity of the decision-making process.

Some committee reports have cited submissions where the majority of sources appear to be AI-generated "hallucinations." This occurs when large language models (LLMs) create content that looks real but does not exist.

Risk of Misinformed Decisions

Christian Downie, a professor at the Australian National University, stated that if this trend continues, there is a significant risk of parliamentarians making decisions based on non-existent evidence. This undermines the purpose of the inquiry process, which is to gather expert and constituent input for better parliamentary decisions.

AI Search Tools Exacerbate the Problem

The situation is complicated by AI services like Google search. Previously, searching for an invented reference would likely show it did not exist. Now, Google's AI summary can create summaries of fake references as if they were genuine. In some cases, Google's AI summary and ChatGPT cite inquiry submissions with fake references as sources, creating a cycle of misinformation.

Impact on Researchers

One submission to an inquiry on family violence and suicide included a hallucinated reference attributed to Divna Haslam, a University of Queensland associate professor. The submission misstated her team's research findings. Haslam described the reference as "scary" due to its realistic appearance, noting that Google's AI summary also summarized it as a real paper. She expressed frustration over inaccurate and inappropriately attributed research.

Submission Author's Response

Drilldown Reports, the author of the submission with the hallucinated reference, confirmed to Guardian Australia that it used AI in its research process. The organization stated it identified the errors in a follow-up submission but missed the deadline for corrections. A spokesperson emphasized the importance of human review but acknowledged that human error occurred in uploading the correct document.