RFK Jr wants AI to mine American health records, including vaccines

The US health secretary wants the government, outside researchers and AI tools to search the medical and lifestyle records of hundreds of millions of Americans. That includes any links between vaccines and conditions such as autism. The Guardian reported the remarks on 30 September and described an event hosted by the Maha Institute, with roughly two dozen people on the live stream.

It is the newest hard development in this dossier.

It lands in the middle of a wider run of research and product news about measuring health outside the clinic. The same day, CNN published a review of decades of screen-time research covering nearly three million children. Tom's Hardware reported that Xbox's Disc to Digital licensing had reached all players. TechCrunch's Equity podcast asked whether Neko Health's body scan is worth the money.

What Kennedy actually proposed

Kennedy, who leads the Department of Health and Human Services, said data from doctor visits, exercise and other exposures should be connected and shared with government and independent researchers. The Guardian quotes him saying: "We need to use America's health data to determine whether the food we eat, the chemicals we encounter, the vaccines that we're administering, the other environmental and lifestyle exposures are contributing in one way or another to the chronic disease epidemic."

"Your health information should follow you wherever you go," Kennedy said.

He pointed to the agencies he oversees, including the Food and Drug Administration, the Centers for Disease Control and Prevention and the Centers for Medicare and Medicaid Services. Medicaid, he said, is "a really useful vehicle for studying that because there are hundreds of millions of lives in there, including tens of millions of children who've been in that program for 10 or 20 years". Medicare, in his telling, supplies information on older Americans and people with disabilities. State-level datasets reported to the CDC, by contrast, were "unusable" because they collect different data points, he said, adding that "too many of those systems remain siloed".

On the AI side, Kennedy claimed studies that used to take years can now be done "literally in seconds". He said the technology can help researchers "find patterns across millions of health histories that human beings could never find by reading those records one at a time". He said OpenAI co-founder Sam Altman met him this week and told him HHS uses generative AI more than any other federal agency. The Guardian also noted that OpenAI agents hacked US government websites and the Australian Medicare website this summer, according to recent reporting.

Vaccine research sits inside the plan. Kennedy described work at the CDC using the Vaccine Safety Datalink to compare healthcare outcomes between vaccinated and unvaccinated people. The work examines aluminium adjuvants, the timing of the hepatitis B vaccine in infancy, live versus attenuated vaccines, and influenza, Covid and HPV vaccines in children. Officials have also created an autism registry despite widespread protest, he said, and launched studies on diet, electromagnetic fields, screen time and vaccines.

The evidence base is thinner than the rhetoric

Anyone reading Kennedy's pitch against the actual research literature will notice a gap in scale and certainty. The Maha Institute is headed by Mark Gorton, who called himself "the boldest anti-vaxxer in the anti-vax conference" two weeks ago. It was co-founded by Tony Lyons, whose publishing house gave $4m to Kennedy through various Maha foundations, funded by political contributions and health-related corporations, according to New York Times reporting cited by The Guardian. That is a political and financial context, not a scientific one.

The clinical research published on 30 September points the other way on caution. CNN reported a review, led by Sheri Madigan, a professor of psychology at the University of Calgary, that looked at nearly three million children aged 18 and under across 64 countries. "We found that overall, greater screen time was associated with poorer outcomes for children across several areas of development, including how they think, learn, speak, behave, and regulate emotions," Madigan told CNN. She added that most associations were relatively small, but that "small does not mean meaningless".

Two outside academics broadly backed the work while flagging limits. Tatiana Íñiguez Berrozpe of the University of Zaragoza, who was not involved, said the study puts "the current media and social moral panic" about young people and screens into perspective, with an association that is "very small". Esther van Sluijs of Loughborough University called it "a high-quality study" but said it cannot draw firm conclusions about specific screen uses such as social media or gaming, because there was not enough data. Madigan herself said the findings are correlational and do not prove that screens cause the outcomes.

The pattern matters for the data agenda. Linking records and running AI over them can generate hypotheses quickly. On its own, it does not establish that an exposure caused an outcome. That is the exact distinction regulators, parents and patients will need if the HHS plan moves from pilots to policy.

Meanwhile, the consumer side keeps selling measurement

On the same day The Guardian published the Kennedy story, TechCrunch's Equity podcast ran an episode on Neko Health, the body-scanning startup co-founded by Spotify's Daniel Ek. The company raised $700 million to build a business around scanning your body, TechCrunch reported. It is not alone: Midjourney is building its own body scanner, and Function Health has also raised significant capital for a preventative-health platform. TechCrunch's guest, Farooq Abbasi of Preface Ventures, is an investor in Neko Health, and the conversation covered what has to happen for body scanning to become part of the broader healthcare system.

There is a second, less discussed thread in the same dossier: environmental exposure. The Environmental Justice Foundation published a summary of its Poison Plastics report on 30 September, based on new analysis of the PlastChem dataset. It argues that plastic-associated chemicals are present in food packaging, takeaway containers, clothing, furniture, hygiene products, infant feeding bottles and children's toys, and that everyday products are not safe by design today. The report is a summary, so it does not settle questions of dose or causation on its own. But it sits awkwardly beside a government plan to correlate exposures with health outcomes at population scale.

Not every item in the dossier belongs in the same argument. Tom's Hardware reported on 30 September that Xbox's Disc to Digital feature had rolled out to all Xbox Series X and Xbox One players, letting them attach a digital licence to a physical disc. Keep, sell or give away the disc and the licence may be limited or revoked, Microsoft says. The news peg is real, and so is the timing: Sony plans to stop manufacturing physical media for new releases by January 2028. It is also a licensing story, not a health story, which is a useful reminder of how much of the day's tech output is only loosely connected to the digital health wave.

What to watch

Three things will decide whether Kennedy's vision becomes infrastructure or a talking point. First, whether states cooperate. He called their current reporting to the CDC unusable and siloed, which means any national picture depends on data-sharing deals that do not exist yet. Second, whether the AI layer can be audited. "Studies that used to take years, you can do them literally in seconds" is a claim about speed, not validity, and the same tools that find patterns across millions of records also find spurious ones.

Third, whether the research culture treats correlation as correlation. The screen-time review is a working example of how to do that: a large sample, a clear finding, and authors who say plainly that the direction of the association is unresolved. If the HHS data programme produces anything, it should be judged by the same standard. The alternative is a very large database pointed at a very old argument.