Report supporting Australia's under-16 social media ban had AI-made errors
What's the story
A report supporting Australia's ban on under-16s using social media has been found to contain errors, possibly created by artificial intelligence, as per The Guardian. The $3.48 million trial, conducted by the UK-based Age Check Certification Scheme (ACCS) last year, tested different technologies for age verification on social media platforms. The report was hailed by Communications Minister Anika Wells as showing "many effective options" for checking people's ages and paving the way for the ban's implementation in December last year.
Investigation details
Errors in report's 'emerging technologies' section
The report, while mostly based on tests of existing age-verification technologies, also features a chapter on "emerging technologies."
This section cites several journal papers about potential new methods for age verification.
However, an ongoing Senate inquiry seeking to further strengthen the social media ban law received a submission highlighting at least two citation errors in this part of the report.
The submission said these "citations ... appear to be AI hallucinated, rather than being based on real sources."
AI usage
ACCS denied using AI initially
When approached by The Guardian Australia, ACCS initially denied any use of AI in the report.
However, they later admitted that AI had been used to rewrite some paragraphs more succinctly.
"We did not use AI in the generation of the report or the cited materials ... Each one of them was checked as genuine links and reports and that they were relevant to the specific issue being cited," an ACCS spokesperson said.
Error revelations
The Guardian Australia's own analysis found errors too
The Guardian Australia's own analysis of the report found six references in the same section with errors.
These included Digital Object Identifiers (DOIs) linking to non-existent papers, incorrect papers, mismatched author name/journal/publication year combinations, and DOI links leading to papers that didn't match the report's claims.
The ACCS spokesperson later provided clarifications and new references for their research, but these too contained errors.
Clarification issues
List of journal articles provided by ACCS had mismatches
The list of journal articles provided by ACCS to correct the errors in the report included mismatched years, authors, and journal titles.
One article ACCS cited was apparently accessed in March 2025, but one of the lead authors of this study confirmed to The Guardian that it wasn't publicly available until June 2025.
Another reference "Weber et al. 2011" couldn't be found in the cited journal, leading to further questions about ACCS's citation practices.
AI admission
AI used to source links, not the report: Spokesperson
The ACCS spokesperson only accepted the use of AI after The Guardian found four links in parts E and K of the report that had metadata showing they were sourced from ChatGPT.
"All of the links were checked and verified and, whilst I apologize if there is an inadvertent error, this was all done by human verification," the ACCS spokesperson said.
Official reaction
Independent senator calls for end of 'slack referencing'
The federal government has been critical of contractors using AI in reports to the government.
Deloitte, for instance, had to refund part of its $440,000 contract after errors were found and admitted to using AI.
In light of these developments, Fatima Payman, an independent senator, said, "The saga of last year's Deloitte AI slop report should have marked the end of slack referencing in reports by government contractors."