AI News AI资讯 5d ago Updated 4d ago 更新于 4天前 46

Report supporting Australia's teen social media ban appears to contain AI hallucinations, Senate hears 支持澳大利亚青少年社交媒体禁令的报告据称包含AI幻觉,参议院获悉

A $3.48 million report underpinning Australia's under-16 social media ban contains multiple fabricated or erroneous academic citations, raising serious questions about the reliability of AI-assisted government research The UK-based Age Check Certification Scheme (ACCS) initially denied using AI in the report, later conceding ChatGPT was used for editing after Guardian analysis found metadata in links revealing ChatGPT as the source Guardian Australia identified at least six citation errors inclu 澳大利亚青少年社交媒体禁令支撑报告被曝存在多处AI幻觉引用错误,包括不存在的DOI和错误的作者信息 报告制作方ACCS最初否认使用AI,后承认使用ChatGPT进行文本编辑,但坚称引用经过人工核查 澳大利亚政府此前已对承包商使用AI生成报告持负面态度,德勤曾因类似问题退还部分合同款项 参议院调查正在审查该报告,政府官员承认难以独立验证引用错误,仅视为"少量错误"

72
Hot 热度
65
Quality 质量
60
Impact 影响力

Analysis 深度分析

TL;DR

  • A $3.48 million report underpinning Australia's under-16 social media ban contains multiple fabricated or erroneous academic citations, raising serious questions about the reliability of AI-assisted government research
  • The UK-based Age Check Certification Scheme (ACCS) initially denied using AI in the report, later conceding ChatGPT was used for editing after Guardian analysis found metadata in links revealing ChatGPT as the source
  • Guardian Australia identified at least six citation errors including non-existent DOIs, incorrect author names, mismatched publication years, and papers that do not support the claims made about them
  • One cited article was reportedly accessed in March 2025 but was not publicly available until June 2025, and its lead author confirmed a cited co-author named "Jamil" never existed on the paper
  • The Australian government has previously penalized contractors for AI-generated errors, with Deloitte refunding part of a $440,000 contract after similar issues surfaced

Why It Matters

This incident highlights a critical vulnerability in AI-assisted policy research: hallucinated citations can lend false credibility to government decisions with real-world consequences, such as regulating an entire generation's access to social media. It also underscores the importance of transparent AI disclosure and rigorous human verification in reports that inform legislation, as unreliable sources can undermine public trust and policy effectiveness.

Technical Details

  • The report in question is a 1,000-page document produced by the UK-based Age Check Certification Scheme (ACCS), with a 26-page citation section covering "emerging technologies" for age verification
  • Citation errors include: DOIs linking to non-existent papers, DOIs pointing to incorrect articles, author/journal/year combinations that do not match any known references, and DOIs linking to papers that do not support the claims made
  • Metadata embedded in at least four hyperlinks within parts E and K of the report explicitly identified ChatGPT as the source, which was the key evidence that forced ACCS to concede AI use
  • ACCS's subsequent clarification response also contained errors, including mismatched years, authors, and journal titles when attempting to correct the original citations
  • One lead author of a cited study confirmed the paper was not publicly accessible until June 2025, yet the report claimed it was accessed in March 2025, and the cited lead author "Jamil" was entirely fabricated

Industry Insight

  • Organizations using AI for research or policy documentation must implement mandatory human verification of all citations and sources, as AI hallucination remains a persistent and well-documented risk that can compromise the integrity of official reports
  • The incident sets a precedent for increased scrutiny of AI-assisted government contracts, suggesting that future procurement may require explicit AI disclosure clauses and independent audit trails for cited sources
  • Policymakers and regulators should treat AI-generated content as inherently untrustworthy until independently verified, and consider establishing standardized verification protocols before relying on AI-augmented research for legislative decisions

TL;DR

  • 澳大利亚青少年社交媒体禁令支撑报告被曝存在多处AI幻觉引用错误,包括不存在的DOI和错误的作者信息
  • 报告制作方ACCS最初否认使用AI,后承认使用ChatGPT进行文本编辑,但坚称引用经过人工核查
  • 澳大利亚政府此前已对承包商使用AI生成报告持负面态度,德勤曾因类似问题退还部分合同款项
  • 参议院调查正在审查该报告,政府官员承认难以独立验证引用错误,仅视为"少量错误"

为什么值得看

该事件揭示了AI辅助生成政府报告时可能产生的严重可信度问题,对政策制定和技术验证领域具有警示意义。报告引用错误直接影响澳大利亚社交媒体禁令的合法性基础,凸显了AI工具在关键决策支持中的风险边界。

技术解析

  • 报告由英国ACCS机构制作,耗资348万澳元,测试多种年龄验证技术以支持澳大利亚16岁以下社交媒体禁令
  • 发现至少6处引用错误,包括DOI链接到不存在的论文、指向错误论文、作者姓名与期刊信息不匹配等问题
  • 部分引用元数据显示来源为ChatGPT,但ACCS声称这是披露AI使用的方式而非错误原因
  • 政府官员承认难以回溯验证链接有效性,仅能确认报告发布时链接有效,后续可能发生变化

行业启示

  • 政府机构在采用AI辅助生成政策报告时需建立严格的引用验证机制,避免AI幻觉影响决策合法性
  • 承包商使用AI工具应透明披露,并建立人工核查流程,特别是涉及学术引用和技术验证的关键内容
  • 政策制定者应审慎对待AI生成内容的可信度,建立独立验证机制以防止技术报告被错误引用损害政策基础

Disclaimer: The above content is generated by AI and is for reference only. 免责声明:以上内容由 AI 生成,仅供参考。

LLM 大模型 Evaluation 评测 Policy 政策 Regulation 监管 Ethics 伦理