In a 92-minute interview with Tucker Carlson on Monday, RFK Jr. drilled down on his vision for the US Department of Health and Human Services (HHS). Artificial intelligence — arguably, a uselessly vague umbrella term — came up multiple times. (As did conspiracy theories and disinformation on vaccines and autism, the medical establishment, and covid-19 deaths.)
As the head of HHS, Kennedy said his federal department is undergoing an “AI revolution.” He implored viewers to “stop trusting the experts,” as highlighted by Gizmodo, and, presumably, put their trust into AI instead of decades of scientific consensus.
He referenced that AI tools were being used to “detect waste, abuse, and fraud” across the federal government — the tagline for Elon Musk’s misguided and disastrous DOGE initiative that’s already led to a scramble to rehire hundreds of wrongfully cut CDC employees. Kennedy also vaguely declared that the CDC will be using AI to “look at the mega data that we have and be able to make really good decisions about interventions,” demonstrating how flimsy his grasp of AI is.
Kennedy said that AI will rapidly accelerate the drug approval process at the FDA, implying it will fully replace animal testing. This is not entirely new, echoing an April announcement from Kennedy’s Food and Drug Administration that the agency will be phasing out animal testing for some pharmaceuticals in favor of “AI-based computational models” and other countries’ safety data. That agency-level change followed the 2022 passage the FDA Modernization Act 2.0 under President Joe Biden, which repealed requirements for all new drugs to undergo animal testing.
There is a lot of ongoing research into the potential for alternate approaches like organ-on-chip systems, organoid cultures, and AI models to supplement or reduce the amount of animal testing used in drug development. And computer modeling has long been a part of pharmaceutical evaluation. However, it’s likely premature to claim that AI can wholly eliminate the need for animal models. “There is currently no full replacement for animal models in biomedical research and drug development,” wrote the National Association for Biomedical Research in an April statement.
Even more concerning were Kennedy’s hints that the current Vaccine Adverse Event Reporting System (VAERS), which is overseen by the CDC, is set to be overhauled and outfitted with AI. (He previously suggested automating the system in April.) VAERS is a first-line detection system for catching rare, previously undetected risks associated with vaccines that has often been misrepresented by anti-vaccine advocates. AI drug testing may sound unsettling, but it would be conducted by external researchers and drug makers. Pharmaceutical companies are incentivized to not release dangerous products because they lose money when they harm people; Kennedy wouldn’t be so directly held to account.
Misinterpretation of VAERS data at the institutional level could sow further distrust in public health and give Kennedy’s newly appointed vaccine advisory committee ammunition to change vaccine recommendations, legitimize their fringe beliefs, and limit vaccine access.
Anyone can report to VAERS (and certain providers are required to report) anytime a person experiences any negative health event in the aftermath of a vaccination. A report to VAERS does not indicate causality. “There’s nothing about VAERS that allows us to determine whether a vaccine caused the reported adverse event,” says Kawsar Talaat, an infectious disease physician and vaccine safety researcher at Johns Hopkins University. “People report things like anger after vaccination,” she says, for which there’s no biologically plausible mechanism relating back to immunization.
Even more serious events, like death following a vaccination, overwhelmingly bear out to be unrelated to the shot itself. “The thing about vaccines is they protect against preventable diseases, not everything else that occurs in life,” says Paul Offit, a vaccine scientist, virologist, and professor of pediatrics at the Children’s Hospital of Philadelphia.
Yet even so, VAERS reports are followed up with CDC investigation through complementary programs like Vaccine Safety Datalink and the Clinical Immunization Safety Assessment Project. The system has worked since its establishment in 1986 to generate hypotheses for potential vaccine side effects and even to detect very rare vaccine risks. For instance, VAERS did successfully pick up the myocarditis associated with mRNA covid-19 vaccines, which only showed up in about one per 30,000 doses, and the blood clotting associated with the Johnson & Johnson covid-19 shot, which affected about one in 250,000 people, Offit notes. “You’re not going to pick that up pre-licensure, so I think VAERS works well,” he says.
“The problem is that anti-vaccine activists use it to mean that anything reported in that system is a real issue, which is obviously wrong,” he adds — echoing Talaat’s point that anyone can report anything.
It’s not clear how Kennedy plans to introduce AI into VAERS, but presumably he means to feed VAERS data into some sort of automated system for identifying alleged vaccine side effects and risks. Earlier this year, the top US vaccine regulator at the FDA was forced out over his refusal to grant Kennedy unfettered access to the VAERS database, out of fears he and his appointees would manipulate the data. Now, with little standing in his way, Kennedy seems poised to do just that.
There is a reasonable argument to be made that the right set of machine learning algorithms or AI tools could streamline the review process for VAERS claims. But AI systems are only as good as their training and parameters. If you feed them faulty information, that’s what they’re going to regurgitate. If you build an AI system to validate your preexisting belief that vaccines are dangerous, that’s exactly what it will do.
Despite the genuine promise that some AI approaches have in health policy and medicine, experts routinely emphasize that we need to tread carefully in building, vetting, and adopting these technologies. Bias, privacy concerns, legal challenges, and user manipulation all remain major issues, according to one 2024 review of 120 studies of generative AI in medicine. (Not to mention hallucinations: In May, the “Make America Healthy Again Commission,” a presidential advisory committee chaired by Kennedy, released a likely AI-generated report containing false citations to studies that did not exist.)
The key question here is if an AI vaccine risk-assessment system could be developed fairly and accurately under Kennedy’s leadership. Offit, at least, doesn’t think so. “Robert F. Kennedy Jr. is an anti-vaccine activist, a science denialist, and a conspiracy theorist,” he says. “He will do everything he can, as long as he is in this position, to make vaccines less available, less affordable, and more feared.”
