US Department Health and Social Care is developing an artificial intelligence tool that will find patterns in data reported to the national vaccine monitoring database and generate hypotheses about the negative effects of vaccines, according to a study list last week published all the artificial intelligence exploit cases the agency had in 2025.
According to the HHS document, the tool has not yet been implemented, and an artificial intelligence inventory report from last year shows that work on it has been ongoing since slow 2023. However, experts fear that the forecasts it generates could be used by Secretary of Health and Human Services Robert F. Kennedy Jr. to implement its anti-vaccination program.
Kenedy, a longtime vaccine critic, changed the childhood vaccination schedule during his term, removing some shots from the list of recommended vaccinations for all children, including against Covid-19, influenza, hepatitis A and B, meningococcal disease, rotavirus and respiratory syncytial virus, or RSV.
Kennedy also called for a change to the current safety monitoring system for collecting data on vaccine harm, known as the Vaccine Adverse Event Reporting System (VAERS), affirmatively that it is withholding information about the true rate of vaccine side effects. He too proposed changes to the federal vaccine injury compensation program that could do this it’s easier for people to sue in the case of adverse events that have not been proven to be related to vaccines.
The VAERS system, managed jointly by the Centers for Disease Control and Prevention and the Food and Drug Administration, was created in 1990 to detect potential safety problems with vaccines after they are approved. Anyone, including healthcare professionals and members of the public, can submit an adverse reaction report to the database. Because these claims are unverified, VAERS data alone cannot be used to determine whether a vaccine caused an adverse event.
“VAERS has always been a hypothesis-generating mechanism at best,” says Paul Offit, a pediatrician and director of the Center for Vaccine Education at the Children’s Hospital of Philadelphia, who previously served on the CDC’s Advisory Council on Immunization Practices. “It’s a noisy system. Anyone can report and there is no control group.”
Offit says the system only shows adverse events that occurred at some point after vaccination; does not prove that the vaccine caused these reactions. CDC your own website states that reporting to VAERS does not mean that the vaccine caused the adverse event. Still, anti-vaccine activists have misused VAERS data for years to argue that vaccines are unsafe.
Leslie Lenert, formerly founding director of CDC’s National Center for Public Health Informatics, says government scientists have been using established natural language processing AI models for several years to find patterns in VAERS data, so it’s no surprise that HHS will move toward adopting more advanced multi-language models.
One major limitation of VAERS is that it does not include data on how many people have received the vaccine, which can make events recorded in the database appear more common than they actually are. For this reason, Lenert says it is vital to combine information from VAERS with other data sources to determine the true risk of an event.
LLMs are also known to produce convincing hallucinations, which highlights the need for people to test any hypotheses generated by LLMs.
“VAERS is intended to be very exploratory. Some people at the FDA are currently treating it as more than just exploratory,” says Lenert, who is currently director of the Center for Biomedical Informatics and Artificial Health Intelligence at Rutgers University.