Federal 'Make America Healthy Again' Report Released with Multiple Reportedly Erroneous and Unverifiable Citations

The federal "Make America Healthy Again" (MAHA) report, released under HHS Secretary Robert F. Kennedy Jr., included hundreds of citations, some of which were reportedly nonexistent or erroneous. Analysts reportedly identified markers consistent with AI-generated text, such as repeated entries, nonexistent studies, and URLs containing "oaicite," suggesting use of tools like ChatGPT.

Risk classification

• Primary risk domain: 3 Misinformation
• Primary risk subdomain: 3.1 False or misleading information

The incident involved ChatGPT generating fictitious scientific studies and citations that were published in a federal report, spreading false information to the public.

Additional risk subdomains

• 5.1 Overreliance and unsafe use: Government officials relied on ChatGPT's outputs to construct a high-stakes federal health report without verifying the accuracy of the generated citations.

Causal factors

• Entity: AI
• Intent: Unintentional
• Timing: Post-deployment

The incident was caused by OpenAI's ChatGPT generating hallucinated, fictitious citations which were then included in a federal health report.

EU AI Act risk tier

• Risk tier: 3 Limited Risk

Risk Level 3: Limited Risk. The system used was ChatGPT, a general-purpose chatbot. Under the EU AI Act, chatbots and generative AI systems that generate text are classified as limited risk, requiring transparency obligations so users know they are interacting with AI-generated content.

AI system and alleged parties

• AI system: ChatGPT (OpenAI)
• AI purpose: Writing Assistant; Technical Text Generation
• Behaviour type: Assistant
• Alleged developer: OpenAI
• Alleged deployer: United States Department of Health and Human Services
• Alleged harmed parties: United States Department of Health and Human Services, Scientific rigor, General public, Academic integrity

Harm severity

Severity ratings are based on this scale.

Highest direct severity in any category: Minor. Severity is scored from Negligible to Catastrophic in each harm category, for harm the reports describe as caused directly or indirectly by the AI system.

• Physical: direct Negligible, indirect Negligible
• Infrastructure: direct Negligible, indirect Negligible
• Property: direct Negligible, indirect Negligible
• Financial: direct Negligible, indirect Negligible
• Environmental: direct Negligible, indirect Negligible
• Malicious content: direct Negligible, indirect Negligible
• Differential treatment: direct Negligible, indirect Negligible
• Civil rights: direct Negligible, indirect Negligible
• Democracy: direct Negligible, indirect Negligible
• Privacy: direct Negligible, indirect Negligible
• Psychological: direct Negligible, indirect Negligible
• Epistemic: direct Minor, indirect Minor
• Child sexual exploitation and abuse: direct Negligible, indirect Negligible

Epistemic

Reported: Yes

Directly caused: ChatGPT generated at least seven entirely fictitious scientific studies and multiple garbled citations, which were published in the official federal report.

Indirectly caused: The publication of these fake citations misled readers and journalists regarding the scientific consensus and evidence base supporting the report's conclusions.

People affected

• Occurrences reported: 1
• People reportedly harmed: 3
• People reportedly exposed: 1000

Potential causes

Management

• Inadequate Quality Control: Management failed to ensure rigorous review before releasing the report.
• Dismissal of AI Risks: Management downplayed AI-generated errors as minor formatting issues.

Technology

• Generative AI Hallucinations: AI model generated fictitious citations and studies that did not exist.
• Unreliable Citation Generation: AI model garbled URLs and fabricated co-authors or journal names.

Data Inputs

• Unvalidated AI Outputs: AI-generated text was directly imported as source data for the report.

Human Factors

• Overreliance on AI Outputs: Authors trusted ChatGPT's citations without checking original sources.
• Lack of Professional Verification: Writers failed to verify the existence of cited studies manually.

Process and Methods

• Absence of Verification Review: No independent review process was used to verify the bibliography.
• Inadequate Drafting Guidelines: No processes existed to regulate or disclose the use of AI in writing.

Regulatory Environment

• Lack of AI Disclosure Standards: No regulatory standards required disclosing AI use in government reports.

Information quality

• Classification confidence: High
• Reason for confidence: The reports provide clear, consistent, and detailed evidence of the AI's role, specifically identifying 'oaicite' markers in the report's URLs and detailing multiple specific instances of fabricated citations. Multiple independent investigations (NOTUS, Washington Post, NYT) confirmed these findings, and the White House subsequently updated the report, validating the claims.
• Ambiguities identified: The exact prompt used to generate the citations and the identity of the specific staff member who ran the queries remain unconfirmed, though the outcome is clear.
• Alternative interpretations: One could argue the errors were purely human formatting mistakes as claimed by the HHS spokesperson, but the technical evidence of 'oaicite' markers strongly refutes this.

An official federal health report from a presidential commission was found to contain fictitious scientific citations generated by OpenAI's ChatGPT. While the report was quickly corrected, the incident highlights the national security risks of government overreliance on generative AI without rigorous human verification, temporarily undermining the credibility of state-level policy outputs.

• Overall national security impact: Minor
• Scope: Single nation
• Primary target: United States
• Alleged perpetrator: Unknown

Threat characteristics

• Imminence: Long-term. The immediate issue was resolved with a corrected report, but the systemic risk of AI overreliance in policy drafting remains an ongoing strategic concern.
• Autonomy: Human-controlled. The AI system acted as a writing assistant, with human operators retaining final control over the publication and dissemination of the report.
• Novelty: Evolved capability. AI hallucination is an established threat, but its occurrence in a high-profile federal policy report represents an escalation of the risk to state operations.

Impact by dimension

• Physical security: Negligible. No kinetic attacks, physical infrastructure damage, or threats to human safety were reported in connection with this incident.
• Information security: Negligible. The incident did not involve adversarial information warfare, intelligence compromise, or hostile foreign propaganda campaigns.
• Sovereignty: Minor. The inclusion of AI-fabricated studies in a presidential commission report temporarily undermined government credibility, but was quickly managed via a corrected publication.
• Economic security: Negligible. No significant threats to financial systems, strategic technologies, or economic stability were identified.
• Societal stability: Negligible. While the incident caused epistemic harm and degraded public trust in federal health guidelines, it did not threaten societal stability or human rights.