这本应是一项简单的任务。美国卫生与公众服务部着手在一起仲裁案件中提交辩方文书，主张仲裁员涉嫌犯错。此类文书经常被提交，在正常情况下，这一切根本不会引起公众关注。但眼下并非正常情况。《华盛顿太阳报》报道称：特朗普政府显然仍未吸取教训，没有学会复核来源。
It seemed like a straightforward task. The Department of Health and Human Services set out to file an argument in an arbitration case, presenting claims about alleged mistakes made by an arbitrator. Such documents are filed all the time, and under normal circumstances, none of this would be notable to the public at all. But these are not normal circumstances. The Washington Sun reported: The Trump administration apparently still hasn’t learned its lesson about double-checking sources.
一家联邦劳工机构在一份近期的仲裁文书中指出了如此多的问题，以至于它暗示卫生与公众服务部使用了人工智能起草该辩方文书。
A federal labor agency flagged so many problems in a recent arbitration filing that it suggested the Department of Health and Human Services used artificial intelligence to draft its brief.
有大量证据支持这些指控：联邦劳资关系管理局在该文书中认定了 10 处“完全捏造”的引用；正如《太阳报》的报道所补充的那样，该部门文书中的 29 个脚注中有 22 个，以及全部 42 处法律或卷宗引用中有 35 处，被认定为“捏造、错误或无关”。还有“错误的页码、案号和其他细节。”
There’s a lot of evidence to support those allegations: The Federal Labor Relations Authority identified 10 “entirely fictitious” citations in the filing; and as the Sun’s report added, 22 of the HHS filing’s 29 footnotes and 35 of the 42 total legal or record citations were deemed “fictitious, erroneous, or irrelevant.”There were also “incorrect page numbers, case numbers and other details.”
特朗普政府拥抱人工智能，有着长期的马虎记录。例如在 7 月，一名国务院代表在艾滋病 2026 大会上做演示时，展示了一张非洲地图，上面多个国家标注错误。路透社报道指出，该图像“带有表明其由 OpenAI 工具生成的人工智能水印”。这很尴尬，但并非孤立事件。例如，边境巡逻官员去年秋天被发现滥用 ChatGPT。大致同一时期，时任国家情报总监的塔尔西·加巴德使用一款 AI 程序决定公开哪些与 1963 年肯尼迪总统遇刺案相关的文件，该过程进展并不顺利。
The Trump administration’s embrace of artificial intelligence has a long track record of sloppiness. In July, for example, a State Department representative delivered a presentation at an AIDS 2026 conference and referred to a map of Africa that mislabeled multiple countries. A Reuters report noted that the image “contained an artificial intelligence watermark that signals it was made with OpenAI tools.”This was embarrassing, but it was not an isolated incident. Border Patrol officials, for example, were caught misusing ChatGPT last fall. Around the same time, Tulsi Gabbard, then the director of national intelligence, used an AI program to determine which documents to release related to the 1963 assassination of President John F. Kennedy, and that process did not go well.
但小罗伯特·F·肯尼迪部长领导的卫生与公众服务部在这方面遭受的羞辱远不止这些。
But Secretary Robert F. Kennedy Jr.’s HHS has suffered more than its share of humiliations on this score.
例如，去年该部门发布了《MAHA报告：让我们的孩子再次健康起来》，在工作人员滥用AI后，该报告引用了多项不存在的科学研究。不久后，食品药品监督管理局推出了一款旨在加快药物审批的AI工具，但该工具并未奏效。
Last year, for example, the department produced “The MAHA Report: Making Our Children Healthy Again,” which cited multiple nonexistent scientific studies after staffers misused AI. Soon after, the Food and Drug Administration unveiled an AI tool intended to speed up drug approvals, but the tool did not work.
就在上个月，卫生与公众服务部（HHS）还引用了不存在的、AI生成的引文，以支持对联邦青少年怀孕预防项目的全面变革。《华盛顿邮报》报道称，一名联邦法官发现，该部门引用了七篇文章，其中两篇完全是捏造的，三篇在其被归属的期刊中根本不存在。如今，这份名单又增加了一些。
As recently as last month, HHS included nonexistent, AI-generated citations to support sweeping changes to federal teen pregnancy prevention programs. The Washington Post reported that a federal judge discovered that the department cited seven articles, two of which were made up entirely and three of which did not exist in the journals to which they were attributed. Now, the list has grown a little longer.
联邦劳资关系管理局在其裁决中提醒卫生与公众服务部：“虽然没有任何规定禁止当事人在准备提交给管理局的文件时使用AI工具，但我们提醒那些这样做的人注意，AI仍然存在重大的错误风险。”
“While nothing prohibits parties from using AI tools when preparing filings for submission to the Authority, we caution those who do so to be aware that AI continues to present significant risks of error,” the Federal Labor Relations Authority reminded HHS in its ruling.
这一警告是值得的，但如果以近期历史为鉴，该部门不太可能把这条提醒放在心上。
The warning was worthwhile, though if recent history is any guide, the department is not likely to take the reminder to heart.