ChatGPT Was Used in the Report That Paved the Way for Australia's Social Media Ban

Key Takeaways

• A Guardian Australia analysis found at least six citation errors in a key report underpinning Australia's under-16 social media ban, including links to academic papers that do not exist and DOIs pointing to incorrect or nonexistent articles.
• The $3.48 million report, produced by UK-based Age Check Certification Scheme (ACCS), was commissioned to test age assurance technologies and was heralded by Australia's Communications Minister as showing "many effective options" for age verification.
• ACCS initially denied using AI in the report, then conceded ChatGPT was used to rewrite paragraphs — after the Guardian identified metadata in four links showing they originated from ChatGPT.
• One study cited in the report was apparently accessed in March 2025, but its lead author confirmed to the Guardian that the article wasn't publicly available until June 2025, that the cited lead author "Jamil" was never on the paper, and that ACCS had summarized the paper incorrectly.

A report that cost $3.48 million and helped pave the way for Australia's under-16 social media ban — one of the most significant internet restrictions any democratic government has implemented — appears to contain citations to academic papers that don't exist, according to a Guardian Australia investigation published August 17th, 2026.

The report, produced by UK-based Age Check Certification Scheme (ACCS), was commissioned to test age assurance technologies as part of Australia's Social Media Minimum Age trial. Communications Minister Anika Wells heralded it as showing "many effective options" for checking people's ages, and it directly informed the ban that came into effect in December 2025. 

The Guardian found six errors in one section of the report: DOIs linking to papers that don't exist, DOIs pointing to incorrect papers, author-journal-year combinations that match no known references, and at least one paper summarized inaccurately.

ACCS initially denied using AI in the report. When the Guardian identified ChatGPT metadata in four links in the document, the organization conceded AI had been used to rewrite paragraphs — but maintained it was not used to generate research or citations, and that all links were checked and verified by humans.

One of the studies cited was apparently accessed in March 2025. Its lead author confirmed to the Guardian that the article wasn't publicly available until June 2025, that the person ACCS cited as the lead author "Jamil" had never been on the paper, and that ACCS had summarized the paper incorrectly. When ACCS provided a correction citing "Weber et al. 2011," the Guardian checked the source ACCS supplied — and found no reference attributed to Weber in it.

Why This Matters Beyond One Contractor's Errors

The technical details of citation errors are important but secondary. What matters here is what the report was used for, and what this reveals about the evidentiary basis on which major internet restrictions are being built.

Australia's social media ban is the most sweeping restriction on children's access to digital platforms that any democracy has enacted. It has been cited by the UK, the EU, and other governments as a model — the same model we've noted in previous coverage that the EU Commission president is expected to invoke in her September State of the Union address. 

The report that helped justify the specific age verification technologies underpinning that ban now contains confirmed citation errors, at least some of which point to papers that do not exist.

The question this raises isn't only about ACCS's research methods. It's about how governments assess the evidence base for major policy decisions. If the report contained AI-hallucinated citations and those errors passed through a government review process without detection — and if the department's response when confronted is that verifying whether the cited papers existed would be "very difficult" — then the quality control infrastructure around evidence-based policymaking is not functioning adequately for decisions of this magnitude.

Professor Christian Downie of the Australian National University, who has previously examined government submissions containing AI hallucinations, told the Guardian that false citations in government reports "could lead to bad decisions, and erode public confidence and trust in institutions."

The Pattern This Fits

We've written extensively about Australia's social media ban — the government's own eSafety Commission found that seven out of ten affected children still have "some access" to banned platforms; Australia subsequently doubled the maximum fine for non-compliant platforms rather than reassessing the policy. 

The ban was presented as an evidence-based intervention backed by research on age assurance technology. Some of that research, it now appears, cited papers that its own authors say weren't accessible when the report was written.

This doesn't invalidate the entire report — the Guardian notes the citation problems are concentrated in a chapter on emerging technologies, and the bulk of the report covers direct testing of existing platforms. But it does matter that the section with hallucinated citations concerns exactly the emerging technologies that would extend the framework beyond its current scope.

It also fits a broader and increasingly important pattern. AI-generated or AI-assisted content is working its way into research, government submissions, and policy documents faster than the verification infrastructure designed to catch errors can keep up. 

Deloitte was required to refund part of a $440,000 government contract last year after similar issues were identified with AI use. The lesson — that contractual arrangements and penalties need to be in place before the report is published, not after — is apparently still being learned.

Australia built a major internet restriction on research that included citations to papers that don't exist. The governments now pointing to Australia as a model should probably know that before they follow it.