My detailed analysis of the documentation underpinning Australia's social media ban for minors has uncovered a systemic problem that calls into question the quality of government decisions made on the basis of "expert" findings. This concerns a thousand-page report prepared by the British organization Age Check Certification Scheme (ACCS) for 3.48 million Australian dollars (~$2.5 million). This document served as the formal justification for the law that came into force on December 10, 2025, restricting access to platforms for users under 16.
Six "hallucinations" in the scientific chapter
The bulk of the report is devoted to practical tests of more than 60 solutions from 48 vendors. However, it is the secondary section, where the authors turn to scientific literature to describe promising developments, that turned out to be a "minefield." Upon verification, I identified at least six violations that fall into four clear categories: fictitious DOIs, links leading to unrelated works, non-existent "author-journal-year" combinations, and correct references attributed with non-existent claims.
For example, one of the cited articles is dated March 2025, although it was actually published only in July. In another case, a reference to a work by Weber et al. 2011 leads nowhere—it is found neither in the specified journal nor in the materials of Monash University that the authors cite. These are not mere typos but classic signs of generative AI "fabricating" sources.
ChatGPT in the metadata: denial and admission
Initially, ACCS representatives categorically denied the use of neural networks, claiming that every reference had undergone manual verification. However, after traces of ChatGPT were discovered in the metadata of four fragments of the report, the position changed. ACCS admitted that AI was used to "condense" certain parts of the text but insisted that the research itself and the final formulations were created without the involvement of OpenAI's chatbot. Moreover, they stated that the ChatGPT metadata itself constitutes "sufficient notice" of AI use. This is, to put it mildly, a debatable claim that does not hold up to scrutiny in terms of academic and governmental transparency.
A "handful of errors" or a systemic failure?
Australia's Department of Infrastructure and Transport acknowledged that it did not verify the contractor's version of the "broken" links, calling such verification "labor-intensive." Officials are trying to chalk it all up to a "handful of errors" across 26 pages of bibliography, but this is a dangerous precedent. Professor Christian Downie from the Australian National University is absolutely right: false references in government documents lead to erroneous decisions and erode trust in institutions. Independent Senator Fatima Payman's demand for a refund and official apologies seems more than justified, especially against the backdrop of the recent Deloitte scandal, which already returned part of its payment for a report with AI errors.
My conclusion: This incident is not just a story about careless contractors. It is a signal that governments worldwide must immediately implement strict source verification standards and contractual sanctions for the use of AI without disclosure. Otherwise, we risk building digital policy on a foundation of neural network "hallucinations" rather than real data.