Federal Judge Blocks HHS Teen Pregnancy Prevention Overhaul as Fabricated Citations Fuel AI Misinformation
A federal judge has temporarily blocked the U.S. Department of Health and Human Services (HHS) from implementing new requirements for federally funded teen pregnancy prevention programs, after concluding that plaintiffs were likely to show the policy was arbitrary and capricious. In an Aug. 19, 2026, memorandum opinion, U.S. District Judge Christopher Cooper said the HHS funding notices cited public health studies that appeared “either not to exist or not to support the propositions for which they are cited,” a problem he described as “a hallmark of AI-generated citations.” The opinion did not establish that HHS used artificial intelligence to prepare the notices, and HHS did not respond to a request for comment from The Washington Post. According to a declaration submitted by the plaintiffs, five of the seven articles cited in the notices could not be located as cited; two appeared to be entirely made up, while three had titles similar to articles published in journals other than those listed by HHS. The case remains in its early stages, and Cooper cautioned that conclusions reached on an expedited record might not become settled findings as the litigation continues. The ruling comes amid growing concern about how official government sources can lend false legitimacy to nonexistent research, particularly when AI search tools treat federal documents as proof that fabricated studies are real.
Congress established the Teen Pregnancy Prevention (TPP) Program in 2010 to fund medically accurate, age-appropriate programs intended to reduce teen pregnancy, with most program funding reserved for replicating programs shown through rigorous evaluation to reduce teen pregnancy or related behavioral risks, and a smaller portion supporting research and demonstration projects to develop and test new approaches. HHS’s Office of Population Affairs posted the new Tier 1 and Tier 2 funding notices on June 23, 2026, and both required recipients to incorporate sexual risk avoidance education, emphasized avoidance or delay of adolescent sexual activity, and promoted “body literacy.” The notices also required recipients to comply with broader HHS priorities and policies involving parental notification; information about potential health risks associated with sexual and reproductive health medications; diversity, equity, and inclusion policies; gender-related policies; and content that “encourages, normalizes, or promotes sexual activity for minors.” The Tier 1 notice separately directed recipients to “affirm marriage and parenthood as meaningful and valued components of adult life.” Three days after issuing the notices, HHS sent termination letters to over 50 recipients of the program’s existing 2023-2028 grant cohort, using substantially identical language stating that the recipients’ projects no longer advanced program goals or agency priorities and that some curricula normalized adolescent sexual activity and were not age-appropriate. Hennepin County, Minn.; King County, Washington; Planned Parenthood of the Heartland; and SIECUS: Sex Ed for Social Change filed suit in July, arguing that HHS had improperly shifted the program away from the evidence-based approaches Congress directed it to fund.
The court’s preliminary ruling was significant because it found the plaintiffs were likely to succeed in showing that the new policy was arbitrary and capricious under the Administrative Procedure Act. The judge said HHS had not adequately explained several aspects of the policy, including its decision to impose abstinence-only and body literacy requirements across the entire program. He also noted that Congress separately appropriated $35 million for sexual risk avoidance education, which strongly suggested that Congress did not intend HHS to apply an abstinence-only requirement throughout the broader TPP Program. The judge did not conclusively decide whether the policy violated the program’s governing statute, nor did he evaluate the plaintiffs’ First Amendment viewpoint-discrimination claim at this preliminary stage. The preliminary injunction prevents HHS from implementing the challenged 2026 policy, including the Tier 1 and Tier 2 funding notices, while the lawsuit proceeds. However, Cooper declined to order HHS to reinstate the terminated grants, citing uncertainty over whether the federal district court, rather than the U.S. Court of Federal Claims, had jurisdiction to provide that relief. As a result, the ruling pauses the policy’s implementation but does not immediately restore funding to organizations whose grants were terminated, leaving many grantees in a difficult position as the legal battle continues.
The citation problems extend far beyond the HHS funding notices themselves, creating what public health experts call a dangerous amplifier for AI-driven health misinformation. Federal health documents are often regarded as authoritative by patients and clinicians, and search engines and AI systems frequently surface them at the top of results. When a nonexistent study is introduced and cited in an official document, that government source can give the citation an appearance of legitimacy it would not otherwise have. That effect is already observable: in a search conducted by the author, Google’s AI Overview presented one of the apparently nonexistent studies cited in the HHS notice as a real article, providing an author, journal, publication year, and a description of the purported research, while citing the federal funding document as its source. Rather than recognizing that the study cannot be verified, the AI-generated response treated the federal document as evidence that the study exists. This illustrates a troubling feedback loop in which an unverified citation in a government document is picked up by an AI tool, repeated as fact, and then further circulated to users who have no way of knowing that the original source never existed.
The problem is particularly acute for clinicians treating children and adolescents, who may be more likely to trust both AI tools and federal sources. Research published in JAMA Pediatrics found that 19.2 percent of adolescents and young adults ages 12-21 had used AI chatbots for mental health advice, and among those users, 63.3 percent said they had not told anyone about that use. Young patients may similarly consult AI about contraception, sexual health, or reproductive development without telling a parent or clinician. Clinicians may therefore encounter patients who cite evidence that appears credible because both an AI system and a federal document present it as legitimate. The problem may be especially difficult to identify when the AI supplies a plausible study title, authors, journal, and findings rather than an obviously false answer. Conflicting information can also damage the clinical relationship: a 2026 study involving 135 adults interacting with a large-language-model-simulated doctor found that disagreement between an AI system and a medical recommendation increased perceptions of medical uncertainty and physician laziness, thereby affecting measures of trust. An AI system’s confidence, however, does not establish that its supporting evidence exists or that its interpretation is accurate, and clinicians cannot assume that patients will understand this distinction.
For telehealth clinicians, these conversations may occur while patients are simultaneously searching online or consulting an AI tool during a virtual visit. Rather than dismissing the information outright, clinicians can ask which tool the patient used, request the original citation, and verify whether the research is available in the identified journal or database. They can also explain why an official source may still contain an unverified citation and distinguish a government policy document from a peer-reviewed clinical study. The episode shows how a fabricated citation in an authoritative source can travel beyond the original publication: once an AI system treats that citation as real, patients may receive misinformation that appears to be supported by both scientific research and the federal government, even when the underlying study does not exist. The legal fight over the TPP program is far from over, and the judge’s decision may ultimately turn on more than the missing citations, but the broader lesson for health care is already clear: official-looking citations are not proof of evidence, AI-generated summaries can amplify phantom research, and clinicians must be prepared to help patients navigate a rapidly changing information environment without dismissing their concerns or damaging the trust that is essential to effective care.