Authors of Australian Social Media Ban Report Admit Using ChatGPT for Editing
The Age Check Certification Scheme (ACCS) admitted to using ChatGPT to edit a key AUD 3.48 million report on age verification technologies for Australia's under-16 social media ban. This admission came after investigators and journalists discovered multiple citation errors, including fake DOIs and fabricated academic references. This incident highlights the risks of AI hallucinations in high-stakes policy-making and could undermine public trust in government decisions. It also underscores the growing scrutiny over government contractors using generative AI tools without proper oversight. Although ACCS claimed AI was only used to make paragraphs more concise, at least six references in one chapter contained errors, such as DOIs linking to non-existent papers or misrepresenting research findings. The agency only admitted to using AI after metadata from ChatGPT was discovered in the report's links.
## BACKGROUND
Age verification technologies, such as biometric facial analysis or digital ID checks, are increasingly evaluated by governments to restrict minors' access to social media. However, when drafting reports on these technologies, the use of Large Language Models (LLMs) like ChatGPT can lead to 'AI hallucinations,' where the system generates plausible-sounding but completely fabricated citations and facts.