A critical report on age verification technology, which the Australian government cited to justify a ban on social media for teenagers, contains references to non-existent scientific publications. This conclusion was drawn by journalists from The Guardian who analyzed the report's content.

The testing was conducted by the British organization Age Check Certification Scheme (ACCS), which spent AUD 3.48 million (approximately USD 2.5 million) on the project. They evaluated over 60 solutions from 48 providers and submitted a final report of about 1,000 pages in ten volumes to the government in August 2025.

At that time, Communications Minister Anika Wells described the findings as evidence that platforms possess "many effective tools" for age verification. The ban on users under 16 years old took effect on December 10, 2025.

Six References with Errors

The bulk of the report is centered around practical tests of existing solutions. Issues arose in the section discussing future developments, where the authors referenced scientific literature.

The first to highlight the discrepancies was an author who submitted a request to a Senate committee currently reviewing amendments to tighten the social media ban for users under 16. The document states that at least two references appear to be typical "AI hallucinations" rather than citations of actual sources.

Journalists from The Guardian confirmed these claims and identified six errors in the same section, categorized into four types:

  • fictitious DOIs;
  • DOIs redirecting to irrelevant works;
  • combinations of "author-journal-year" not corroborated by any publication;
  • correct links to articles, but lacking the claims attributed to them by the report.

For instance, one article was dated by ACCS as March 2025 but was actually published only in July. In another case, a reference to Weber et al. 2011 led nowhere; it was not found in the cited journal nor in the study from Monash University, as indicated by the report's authors.

ChatGPT in Metadata

An ACCS representative firmly denied the use of AI in the preparation of the document. He stated that the organization did not apply neural networks to the text or the cited materials, and that each reference was thoroughly checked for accuracy and relevance.

However, after The Guardian discovered four references in some sections of the report with metadata indicating the involvement of ChatGPT, the organization's stance had to be softened. ACCS acknowledged that artificial intelligence did rewrite certain segments for brevity, but insisted that the research and the final text were created without the involvement of OpenAI's chatbot.

Representatives from the organization also mentioned that the ChatGPT metadata alone serves as sufficient notification of AI usage, and since the references lead to the appropriate sources, there should be no complaints.

Nevertheless, a review of the revised list of sources provided by ACCS revealed new inconsistencies: years, authors, and journal titles did not match the original citations from the report.

A "Handful of Errors" in a Thousand Pages

The Department of Infrastructure and Communications stated that they are investigating the claims and will contact the contractor. During Senate hearings on August 14, officials relayed ACCS's version: they claimed the links were functional at the time of publication but "broke" later.

First Assistant Secretary Sarah Vandenbroek clarified that the department did not independently verify this claim, deeming such verification "labor-intensive." However, the department emphasized that it concerns only a "handful of errors" among 26 pages of bibliography.

Christian Downey, a professor at the Australian National University, warned that false references in government documents could lead to erroneous decisions and undermine trust in institutions. He argued that this situation necessitates the establishment of clear requirements in contracts and the introduction of penalties for violations.

Independent Senator Fatima Payman called for an official apology from the government and a refund from the contractor. She reminded that a similar incident with Deloitte a year ago should have taught the authorities to be more careful with sources. In October 2025, the consulting firm had already returned part of the payments from a AUD 440,000 contract, admitting to using AI in the preparation of a flawed report.

It is worth noting that in July, the Australian regulator eSafety filed a lawsuit against Telegram, accusing the messaging service of failing to remove materials promoting terrorism.