The claims in the speech: a line-by-line audit

On 17 September 2026 the US Health Secretary made a series of specific, checkable factual claims to the Children's Health Defense conference. Several are already circulating in Canada. Here is what each one rests on.

A QuackeryWatch audit · Posted 18 September 2026 · Companion to the RFK Jr. running file

1. "Only one of the 94 doses was placebo-controlled"

The claim. Of the 94 doses of 18 vaccines on the childhood schedule, only the COVID-19 vaccine was subject to a placebo-controlled trial.

This one is simply false, and it fails on the very example he chose to illustrate it. The National Academies have documented more than fifty placebo-controlled trials of plasma-derived hepatitis B vaccine alone, involving over a hundred thousand participants. Pneumococcal vaccine was tested by randomising 17,437 Gambian infants to vaccine or a placebo containing no pneumococcal antigens, with roughly two years of follow-up; it showed 77% efficacy against invasive disease and a 16% reduction in deaths from all causes, with fewer serious adverse events in the vaccinated group than the placebo group.

Polio, chickenpox and rotavirus vaccines were also originally tested against placebos.

FactCheck.org, reviewing the same speech, identifies the mechanism that keeps this claim alive: it depends on moving the goalposts about what counts as a placebo — typically by insisting that only saline qualifies, which disqualifies trials that used another vaccine or the adjuvant alone as the comparator. It is also not always ethical to withhold a proven vaccine from children in order to run one.

2. The hepatitis B slide: 147 infants, five days, no placebo

The claim. A slide of the manufacturer's package insert, with the assertion that the pre-licensure trial enrolled 147 infants, monitored them for five days, and used no placebo — inadequate for a product given to millions of children.

The number is real. The description of what it is, is not. That trial was a bridging study for a newer recombinant formulation, designed to demonstrate equivalence to hepatitis B vaccines that were already licensed and already tested. Bridging studies are small and short by design, because they are not the safety evidence — they are the link to it.

The evidence underneath includes the fifty-plus placebo-controlled trials noted above, and the first recombinant hepatitis B vaccine was studied at approval in at least 717 children and adults across at least six trials, each running six months or longer.

3. The package insert adverse-events list

The claim. A second slide listed the adverse events from the insert — anaphylaxis, arthritis, lupus, Guillain-Barré syndrome, transverse myelitis, Stevens-Johnson syndrome, psychiatric and eye disorders — presented as the illnesses of the chronic disease epidemic, with 454 such events across all 18 vaccines.

This is a category error, and it is the most reusable claim in the speech, because anyone with an internet connection can download an insert and repeat it.

The adverse-events section of a package insert is a log of events reported to the manufacturer after marketing, by anyone, in any temporal proximity to the product. Inclusion is not a finding of causation, and the inserts themselves say so in the surrounding text — typically noting that because the reports are voluntary and from a population of uncertain size, it is not possible to establish a causal relationship. It is a filing cabinet, not a conclusion. Reading it as a list of things the vaccine does is like reading a hospital's complaints inbox as a list of its diagnoses.

4. "Fewer than 1% of vaccine injuries are reported"

The claim. That the CDC's own 2010 study — he named it the Lazarus study — found fewer than 1% of vaccine injuries are reported, meaning over 99% are missed.

The sentence exists. It is not a finding of that study. It appears in the introductory background section of the ESP:VAERS report, where the authors are summarising prior literature on adverse-event reporting generally — in the same passage that discusses under-reporting of adverse drug events to the FDA. It is a preamble, not a result.

What the project actually did was build an automated detection system at a large practice: 1.4 million doses of 45 vaccines given to 376,452 people, from which it flagged 35,570 possible reactions, or about 2.6% of vaccinations. "Possible reactions" there includes minor, expected effects such as soreness and fever. And the comparison the claim depends on — how many of those were reported to VAERS — was never completed and published.

One of the report's own authors has since explained the figure to FactCheck.org: it reflects the fact that many vaccine side effects are mild and expected, and so not worth reporting. And separate research has found that the system identifies serious adverse events considerably more completely than minor ones — which is the opposite of the implication the claim carries.

The honest position is the boring one. VAERS is a passive system, under-reporting is real and has never been precisely quantified, and it also contains reports of events that had nothing to do with a vaccine. It under-reports and over-reports at once, which is why active surveillance systems exist alongside it.

5. The autism comparison: 0.8 per 10,000 in 1970, one in 31 today

The claim. A 1970 Wisconsin study of 900,000 children found fewer than one case per 10,000; today the figure is one in 31 nationally, one in 19 in California, one in 12.5 among boys. Researchers then knew what they were looking for, so the rise must be real.

The two numbers are not measuring the same thing, and could not be. In 1970 autism had no distinct diagnostic category at all: it entered the DSM as a separate disorder only in 1980. The 1987 revision added pervasive developmental disorder not otherwise specified and dropped the requirement that onset occur in infancy. The 1994 edition introduced the spectrum concept and added Asperger's disorder, making a large population diagnostically visible for the first time. In 1991 US federal education law made autism its own special-education category, creating an administrative reason to record it. In 2013 the DSM consolidated the subcategories into a single spectrum diagnosis.

The modern figure comes from surveillance that counts a child if records show a written diagnosis, a special-education classification, or a billing code. The 1970 figure counted children meeting a narrow clinical description that did not yet formally exist. Comparing them is not a measurement of autism; it is a measurement of the definition.

6. The Guinea-Bissau DTP study

The claim. That Danish researchers in Guinea-Bissau found girls who received the DTP vaccine were ten times more likely to die over the following three months than unvaccinated children, dying of common infections nobody had connected to the vaccine — and concluded the vaccine was killing more children than the diseases it prevented.

The study is real, published in 2017, and the finding on all-cause mortality is as described. What the speech omits is what kind of study it is and what happened to it afterwards.

It is observational. The comparison arises because some infants were vaccinated at a three-month visit and others not until six months, which the authors treated as a natural experiment — not a randomised allocation. Children who arrive late to a clinic differ from children who arrive on time in ways that also affect survival. The finding has been examined and disputed at length in the years since, and it has not displaced the mainstream position on DTP.

One disclosure the audience did not receive: one of the researchers on that Danish team now consults for the reconstituted vaccine advisory committee that voted to end the universal hepatitis B birth dose, and the team has since been awarded a US no-bid contract to study hepatitis B vaccination in newborns in the same country.

7. "The Institute of Medicine ordered the science and it was never done"

The claim. That the Institute of Medicine reviewed 158 injuries, found clear evidence for some, clear absence for a few, and no science at all for most — and that thirteen years later none of it has been done.

"Insufficient evidence to accept or reject" is not "no science exists." It is a formal conclusion meaning the available studies were not adequate to settle that particular pairing — often because the outcome is very rare. It is a real and fair criticism of the evidence base, and the Institute did call for more research. It is not a finding that the question was never studied, and it does not transfer to the schedule as a whole: the Institute's separate review of the childhood immunization schedule itself found no evidence that the schedule is unsafe.

8. Worth recording: chemtrails

Between the section on electromagnetic fields and the section on vaccines, the Secretary told the audience that his department is also looking at chemtrails and the health impact of aircraft emissions. It passed without comment in the room and has gone largely unreported. We note it here because a health department examining chemtrails is itself a documentable fact about the department.

9. Shorter items

"Ninety-four doses of 18 vaccines." Physicians told ABC News that, setting aside annual influenza and COVID-19 shots, children generally receive roughly 30 vaccine doses before the age of 18, many of them in combined injections. The higher figure counts every annual dose across eighteen years.

The UC Davis study. He cited research led by Irva Hertz-Picciotto at the UC Davis MIND Institute as showing the autism epidemic is real. The 2009 study found that about a third of the increase in California diagnoses was attributable to changes in diagnostic criteria, and did not examine the other factors that affect recognition. The MIND Institute stated publicly in September 2025 that the study did not indicate autism was an epidemic.

"No 70-year-old with profound autism." Offered as proof the rise is real. Autism researchers point out that older adults are far less likely to carry a diagnosis, having grown up before the diagnosis existed, while still meeting the criteria for it. Not being diagnosed and not having the condition are different things — which is the entire point of the diagnostic-criteria problem above.

Thimerosal. He described it as devastating children's brains. Multiple studies have evaluated it; there is no evidence of harm at the doses used. It was removed from childhood vaccines in the United States more than twenty years ago as a precaution, and it contains a different and less toxic form of mercury than the kind that accumulates in fish.

Obesity down 7.3%. Gallup found a decline of about 7.8% between 2022 and 2026. Three-quarters of it occurred between 2022 and 2024, before the current administration took office.

"Real food sales went up by 8%." This traces to a financial-industry analysis of a handful of categories — yogurt, nut butters, frozen foods — finding that less-processed products grew faster than highly processed ones within those categories. It does not support a claim about what Americans are eating overall.

Why a Canadian site is auditing an American speech

Because the claims travel and the corrections do not.

Every item above will reach Canadian parents stripped of context, as a slide, a screenshot or a clip — and each arrives with the authority of a serving national health minister. A Canadian physician, midwife or public health nurse who has thirty seconds to respond cannot relitigate the hepatitis B licensure record from memory.

What they can do is ask the single question that unlocks nearly every item on this page: what sits just outside the frame? The bridging study has fifty trials behind it. The insert list has a causation disclaimer beside it. The 1970 prevalence figure has four revisions of the diagnostic manual between it and today. The 82% miscarriage rate has seven hundred women missing from underneath it.

The number is almost always real. The picture around it is almost always cropped.