An Australian government report on age verification technology was found to contain fictitious references, raising concerns about AI use in its preparation.
A recent analysis has revealed significant inaccuracies in an Australian government report on age verification technology, which was utilized to justify a ban on social media access for teenagers. The Guardian's investigation highlighted that the report contains references to non-existent scientific publications.
The report, prepared by the Age Check Certification Scheme (ACCS), was commissioned by the Australian government and involved a budget of AUD 3.48 million (around USD 2.5 million). It assessed over 60 age verification solutions from 48 providers, culminating in a comprehensive document of nearly 1,000 pages submitted in August 2025. Communications Minister Anika Wells had previously characterized the findings as proof that social media platforms have "many effective tools" for age verification, leading to a ban on users under 16 years effective December 10, 2025.
Discrepancies in References
The report's section on future developments was found to be particularly problematic, as it included what were later identified as "AI hallucinations"—references that do not correspond to actual publications. The initial discovery was made by an author who approached a Senate committee currently reviewing amendments to the social media ban.
Following this, The Guardian identified six specific errors in the document's references, which fell into four categories: fictitious Digital Object Identifiers (DOIs), DOIs that led to unrelated works, incorrect author-journal-year combinations, and correct links that did not support the claims made in the report. For example, an article attributed to March 2025 was actually published in July, and a reference to Weber et al. 2011 was untraceable in the cited journal.
AI Involvement Controversy
Despite initial denials from ACCS regarding the use of AI in the report, metadata from some sections indicated that ChatGPT was involved in rewriting parts of the document. An ACCS spokesperson clarified that while AI was used for brevity, the core research and text were created without OpenAI's chatbot. They asserted that the metadata serves as a clear indication of AI usage and claimed that the references led to appropriate sources.
However, a revised list of sources from ACCS continued to show discrepancies in years, authors, and journal titles compared to the original citations, raising further questions about the report's accuracy.
Government Response and Implications
The Department of Infrastructure and Communications announced an investigation into the claims and plans to communicate with the contractor. During Senate hearings, officials indicated that the links were functional upon publication but may have broken later, a claim that was not independently verified due to the labor-intensive nature of such checks. The department maintained that the issues were limited to a "handful of errors" within the bibliography of 26 pages.
Christian Downey, a professor at the Australian National University, expressed concern that such inaccuracies could result in misguided decisions and erode public trust in governmental institutions. He emphasized the need for clearer contractual requirements and penalties for non-compliance. Independent Senator Fatima Payman has called for an official apology from the government and a reimbursement from the contractor, recalling a similar incident involving Deloitte a year prior, where the firm returned part of a AUD 440,000 contract payment after admitting to AI use in a flawed report.
This report emerges amidst ongoing scrutiny of digital platforms, as in July, Australia’s eSafety regulator had filed a lawsuit against Telegram for not removing content that promotes terrorism.
