A key document that Australian authorities used as justification for banning social media for teenagers turned out to be riddled with traces of artificial intelligence. This refers to a report by the British organization Age Check Certification Scheme (ACCS) on testing age verification technologies. As a result of my review, I found that the document contains links to non-existent scientific publications — a classic sign of neural network "hallucinations."
The trials cost 3.48 million Australian dollars (~$2.5 million). ACCS specialists tested more than 60 solutions from 48 vendors and in August 2025 handed the government a final report of about a thousand pages in ten volumes. Communications Minister Anika Wells then stated that the data confirmed the existence of "many effective tools" for age verification. The ban itself for users under 16 took effect on December 10, 2025.
Systematic errors in the scientific chapter
The main part of the report is devoted to practical trials of existing solutions, and there are no problems there. However, in the chapter on promising developments, where the authors turn to scientific literature, serious discrepancies came to light. I counted six violations that fall into four categories: fictitious DOIs, DOIs leading to unrelated works, "author-journal-year" combinations not confirmed by any publication, and correct references but without the claims the report attributes to them.
For example, ACCS dated one article to March 2025, although it was actually published only in July. In another case, a reference to Weber et al. 2011 led nowhere — it was not found either in the specified journal or in the Monash University study to which the authors referred.
ChatGPT in the metadata
ACCS representatives initially categorically denied the use of AI, claiming that every reference had been carefully verified. However, after four references with metadata directly indicating ChatGPT were found in some parts of the report, the position had to be softened. ACCS admitted that the neural network had rewritten individual fragments for brevity, but insisted that the research itself and the final text were created without the involvement of OpenAI's chatbot.
Notably, a check of the corrected list of sources, which ACCS additionally provided, revealed new inconsistencies: years, authors, and journal names diverged from the original citations in the report.
A "handful of errors" or a systemic problem?
The Department of Infrastructure and Transport said they are reviewing the complaints. At Senate hearings, officials relayed ACCS's version: the references were working at the time of publication and "broke" later. First Assistant Secretary Sarah Vandenbroek admitted that the department had not independently verified this version, considering the check "labor-intensive." At the same time, the department emphasized that this concerns only a "handful of errors" across 26 pages of bibliography.
Professor Christian Downie of the Australian National University warned that false references in government documents lead to erroneous decisions and destroy trust in institutions. Independent Senator Fatima Payman demanded an official apology and a refund from the contractor, recalling the recent incident with Deloitte, which in October 2025 returned part of its payments under a contract worth 440,000 Australian dollars for using AI in a report with errors.
My analysis: This situation is a warning sign for the entire industry. When government decisions affecting millions of citizens are based on data that has not undergone even basic verification for reliability, we are witnessing not just negligence but a systemic failure in governance. Senator Payman's demand for a refund is a minimal measure. It is critically important to enshrine in contracts with contractors legal liability for the use of AI and mandatory validation of every source. Otherwise, neural network "hallucinations" could underpin not only reports but also laws.