The Department of Health and Human Services is creating an artificial intelligence tool to generate hypotheses about vaccine harm claims

US Ministry The Department of Health and Human Services is developing a generative artificial intelligence tool to find patterns across data reported to a national vaccine monitoring database and to generate hypotheses about adverse effects of vaccines, according to an inventory released last week of all the use cases the agency had for artificial intelligence in 2025.

The tool has not yet been deployed, according to a HHS document, and an AI inventory report from the previous year shows it has been in development since late 2023. But experts worry that the predictions it generates could be used by HHS Secretary Robert Kennedy Jr. to advance his anti-vaccine agenda.

Kennedy, a longtime critic of vaccines, overhauled the childhood vaccination schedule during his year in office, removing several doses from the list of recommended vaccinations for all children, including vaccines for COVID-19, influenza, hepatitis A and B, meningococcal disease, rotavirus, and respiratory syncytial virus.

Kennedy also called for reform of the current safety monitoring system for collecting vaccine injury data, known as the Vaccine Adverse Event Reporting System, or VAERS, claiming that it obscures information about the true rate of side effects from vaccines. He has also proposed changes to the federal Vaccine Injury Compensation Program, which could make it easier for people to sue for adverse events not proven to be linked to vaccines.

VAERS, which is jointly administered by the Centers for Disease Control and Prevention and the Food and Drug Administration, was created in 1990 as a way to detect potential safety issues in vaccines after they are approved. Anyone, including healthcare providers and members of the public, can submit a report of an adverse reaction to the database. Because these claims have not been verified, VAERS data alone cannot be used to determine whether a vaccine caused an adverse event.

“VAERS has always been, at best, a hypothesis-generating mechanism,” says Paul Offit, MD, a pediatrician and director of the Vaccine Education Center at Children’s Hospital of Philadelphia, who was previously a member of the Centers for Disease Control and Prevention’s Advisory Board on Immunization Practices. “It’s a noisy system. Anyone can report, and there’s no monitoring group.”

Offit says the system only shows adverse events that occurred post-immunization; It is not proven that the vaccine caused these reactions. The CDC’s website says that submitting a report to VAERS does not mean a vaccine caused an adverse event. Despite this, anti-vaccine activists have misused VAERS data over the years to argue that vaccines are not safe.

Government scientists have been using traditional natural language processing AI models to look for patterns in VAERS data for several years, so it’s not surprising that the Department of Health and Human Services is moving toward adopting more advanced large-scale language models, says Leslie Lehnert, formerly founding director of the Centers for Disease Control and Prevention’s National Center for Public Health Information.

A major limitation of VAERS is that it does not include data on the number of people who have received the vaccine, which may make events recorded in the database appear more common than they actually are. For this reason, Lehnert says it is important to correlate information from VAERS with other data sources to determine the true risk of an event.

LLMs are also known to be good at producing convincing hallucinations, underscoring the need for humans to follow any hypotheses generated by LLMs.

“VAERS is supposed to be very exploratory,” says Lehnert, who is currently director of the Center for Biomedical Informatics and Health AI at Rutgers University. “Some people at the FDA are now treating it as more than just exploratory.”