SYDNEY  –  A government-commissioned report on Australia’s under-16 social media ban contained fabricated citations, an analysis finds. A $3.48m report that helped clear the way for Australia’s under-16 social media ban has come under scrutiny after researchers used ChatGPT to edit sections later found to contain fabricated citations. The Age Check Certification Scheme (ACCS), which ran the age assurance technology trial, initially denied using AI before conceding it had rewritten paragraphs with the tool. Guardian Australia’s analysis identified six flawed references in the report’s “emerging technologies” chapter. Some Digital Object Identifiers (DOIs) were linked to non-existent research papers; some of the DOIs indicated papers with authors, journals, and years that nowhere else existed in academic literature; and one DOI led to a paper which did not say what was reported by the paper. A spokesperson for ACCS, in a conversation with Guardian Australia, firstly said that no artificial intelligence was used to produce the report and secondly admitted that ChatGPT metadata could be found in four links in two parts of the report. It argued that the inclusion of metadata amounted to disclosure and each and every link was manually verified. Attempts to correct the flagged citations produced a fresh batch of errors, including one paper an ACCS source claimed was accessed in March 2025, months before its actual June 2025 publication date, according to the article’s lead author. Communications Minister Anika Wells was laudatory about the report, pointing out many good solutions regarding “age verification”. The ministerial department informed a Senate inquiry that it met with ACCS, which explained the issue with faulty links but not fabrication.  The Australian National University’s Christian Downie said false references, whether AI-generated or not, increase chances of bad policy-making and undermine public confidence. Independent Senator Fatima Payman likened this situation to a Deloitte report issued last year in which Deloitte refunded a portion of a $440,000 government contract for faulty AI-generated references.