AI-Generated Citations Undermine Report Supporting Australia's Teen Social Media Ban

A $3.48 million report that helped justify Australia's ban on social media for under-16s contains fabricated academic citations, according to an investigation by The Guardian. The report, produced by the UK-based Age Check Certification Scheme last year, tested various age-verification technologies that could enforce the ban. Communications Minister Anika Wells cited it as evidence of "many effective options" for age-checking, and the report's findings helped pave the way for the ban to take effect in December. But when the authors were asked this month about citation errors flagged in a Senate inquiry submission, they initially denied using artificial intelligence at all—then reversed course, admitting ChatGPT had been used to rewrite paragraphs for brevity.

The problems run deeper than a simple editing assist. Guardian Australia's analysis of the report identified at least six faulty references in a chapter on emerging age-verification technologies. Some citations point to Digital Object Identifiers—the unique codes that link to specific academic papers—that lead nowhere. Others link to papers that don't exist. Some combine author names, journal titles, and publication years in ways that match no known research. One paper the ACCS cited as accessed in March 2025 wasn't actually published until June 2025, according to one of its lead authors. That same author confirmed the person ACCS credited as the paper's lead—listed only as "Jamil"—had never been on the paper at all, and that ACCS had misrepresented what the research actually said.

When asked to clarify, ACCS provided a corrected list of references. That list too contained errors. A citation attributed to "Weber et al. 2011" could not be found in the journal ACCS claimed it came from. When pressed, ACCS said the reference came from a Monash University study they provided to the Guardian—but the Monash paper contained no reference to Weber. Only after the Guardian identified ChatGPT metadata embedded in four of the report's links did ACCS concede the AI tool had been used. The organization maintained it had disclosed the use by including the metadata, and argued that because the links themselves were correct, there was no real problem. "All of the links were checked and verified," a spokesperson said, "and whilst I apologise if there is an inadvertent error, this was all done by human verification."

The government has shown little patience for this kind of sloppiness before. Deloitte refunded part of a $440,000 contract last year after admitting it had used AI to generate parts of a report and errors were discovered. Yet when department officials met with ACCS, they accepted the contractor's explanation that the broken links were simply the result of URLs changing over time—without independently verifying the claim. Sarah Vandenbroek, the department's first assistant secretary, told a Senate hearing on Friday that ACCS had assured them the links all worked when the document was published. "Something may have changed in the interim," she said, adding that while some links may be broken, "the source documents do still exist, they are still valid." The officials acknowledged it would be "very difficult" to go back and check.

Christian Downie, a professor at the Australian National University's school of regulation and global governance who has studied AI hallucinations in government submissions, warned that false citations—whether generated by AI or not—can lead to flawed policy decisions and corrode public trust in institutions. "One would assume the government is already thinking about this," he said, "and that is going to have to require contractual arrangements and presumably penalties to discourage this type of behaviour." Independent Senator Fatima Payman was blunter. "The saga of last year's Deloitte AI slop report should have marked the end of slack referencing in reports by government contractors," she said, calling on the government to demand an apology and refund from ACCS. The question now is whether the government will treat this as a serious breach of trust, or whether it will accept the contractor's assurances and move on.