A Federal Health Report Fabricated Citations and the White House Called It Formatting
What happened
In May 2025, the White House released a report on the decline in Americans' life expectancy, positioned as a serious piece of health policy research from the Make America Healthy Again commission. Within days of publication, independent scientists and journalists had found something that undercut the entire document: a pattern of citation errors, studies that appeared to have been fabricated outright, and formatting artifacts consistent with content generated by an AI language model. The report had been assembled in just over three months, and it showed.
The problems were not minor editorial slips. Researchers checking the report's references found citations to studies that do not exist in any scientific database, blended real and invented sources, and passages where the writing itself bore the hallmarks of generative AI output. AI language models are known to produce plausible-sounding but entirely fictitious academic references, a behavior researchers call hallucination. When those references are dropped into a government health report and published without verification, the fabrications travel with the authority of official policy behind them.
Experts and analysts reviewing the document pointed to clear signs that AI tools, likely a large language model such as ChatGPT, had been used to generate or supplement the report's citations and supporting text. A three-month turnaround for a document of this scope left little room for the kind of citation-by-citation verification that academic and government publishing ordinarily requires. The evidence suggests the generated text was accepted without being checked against the underlying sources it claimed to represent.
The White House and the Department of Health and Human Services, when pressed, attributed the errors to "formatting issues" rather than acknowledging the role AI tools may have played in the report's production. That framing recast a credibility problem as a technical one. Independent experts declined to accept that characterization, and journalists published detailed breakdowns of specific fabricated references that could not be explained by formatting alone. The gap between the official explanation and what reviewers found remains unresolved.
What the episode reveals is not simply that AI can produce errors, but that government publishing pipelines have no mechanism to detect them before release. A report goes out under the seal of the White House and the authority of the Department of Health and Human Services, and there is currently no requirement to document which parts of it were generated, which were verified against primary sources, and who signed off on the distinction. Without a provable record of what a system produced and what a human confirmed, the official response is left to characterize hallucinated citations as formatting problems, and the public has no independent basis to check that claim.
Reported impact
- Affected parties
- Not publicly disclosed
- Harm type
- Not publicly disclosed
- Scale
- Not publicly disclosed
- Financial impact
- Not publicly disclosed
- Regulatory action
- Not publicly disclosed
Classification
Relevant governance controls
Governance control mapping is not available for this record.
- No controls mapped
Not publicly disclosed
Control mapping is analytical. It does not state that any control would have prevented the incident.
Sources and evidence
This record was researched and written by the Index. The event is also catalogued in the following database, which is listed for cross-reference.