The $3.48 Million Report Behind Australia's Social Media Ban Has Citations That Don't Exist

The age assurance technology trial report that the Albanese government relied on to justify Australia's under-16 social media ban contains references to academic articles that do not exist, with at least some appearing to be AI-generated fabrications, according to a Guardian Australia analysis published on August 17, 2026 (Guardian).

The report was produced by the UK-based Age Check Certification Scheme (ACCS) at a cost of $3.48 million. Communications minister Anika Wells hailed it as evidence of "many effective options" for verifying ages online, and it cleared the path for the ban to take effect on December 9, 2025, making Australia the first country to enforce such a restriction (Reuters). Reuters reported that 86% of Australians aged eight to 15 used social media.

The problems are extensive. Guardian Australia's own analysis of the ACCS report identified six references in the emerging technologies section containing errors. A submission to an ongoing Senate inquiry flagged at least two citation errors and stated the references "appear to be AI hallucinated, rather than being based on real sources" (Guardian).

AI hallucination is what happens when a language model like ChatGPT generates text that sounds plausible but is simply made up — invented author names, fake journal titles, nonexistent papers that look real at a glance. The errors in the ACCS report take multiple forms. DOIs (digital object identifiers — the unique addresses assigned to academic papers) in the report link to papers that do not exist. Other DOIs point to real papers, but not the ones the report claims to be citing. Author names, journal titles, and publication years are combined in ways that match no known references. In some cases, the DOI link leads to a genuine paper that does not say what the ACCS report attributes to it.

One cited study was apparently accessed in March 2025, but the lead author confirmed to the Guardian that the paper was not publicly available until it was published in June 2025. The same author said "Jamil", whom ACCS listed as the paper's lead author, had never been on the paper, and that the ACCS report had summarised the work incorrectly. Another reference, "Weber et al. 2011", could not be found in the journal cited. An ACCS spokesperson said the reference came from a Monash University study, but the Monash paper ACCS supplied contained no reference attributed to Weber.

Then it got worse for ACCS. The spokesperson initially denied any use of AI in producing the report. That position became untenable when the Guardian identified four links in parts E and K of the report containing metadata identifying ChatGPT as their source. Only then did the spokesperson concede AI had been used, though the admission was narrow: AI was used to rewrite paragraphs more succinctly. The spokesperson denied the citation errors were caused by AI hallucinations.

ACCS also argued that embedding ChatGPT metadata in the links amounted to disclosure of AI use, and since the links themselves were correct, there was no problem. That is a generous reading of what constitutes transparency. A reader encountering a hyperlink with ChatGPT metadata embedded in it would need to inspect the underlying metadata to detect it, which is not the same as a transparent declaration of AI assistance in the report's methodology or acknowledgements.

The attempted corrections compounded the damage. References ACCS supplied to fix the problematic citations also contained errors, with years, authors, and journal titles that did not match the original citations. For a report commissioned to establish the reliability of age assurance technologies, the inability to reliably cite its own sources is a pointed irony.

The broader context matters here. The trial, evaluated by KJR (KJR), assessed age verification, age estimation, age inference, successive validation, parental control, and parental consent across multiple technologies. The final report, published by the Department of Infrastructure in August 2025 (Infrastructure.gov.au), found that some errors in age assurance are inevitable (Guardian, August 2025). The government pressed ahead regardless, announcing it would introduce age restrictions following the report's release (Biometric Update).

What is now in question is not just the provenance of a single document but the evidentiary chain behind a world-first piece of legislation. The ACCS report was the technical foundation ministers pointed to when arguing the ban was workable. If that foundation contains fabricated citations and undisclosed AI assistance, the political and legal scrutiny now falling on it through the Senate inquiry is both predictable and overdue.

The ACCS trial has formally concluded and is no longer accepting new entries (ageassurance.com.au). The ban it underpins has been in force since December. Parliament does not get to retroactively un-ring this bell, but the Senate inquiry now has a straight question to answer: whether the government legislated a world-first restriction on the basis of a report whose own authors could not keep their references straight.