Artificial Intelligence Lies about Hanta Virus
Recently Hanta Virus has made the news, and the buzz is that there may have been human to human transmission on board a cruise ship. This led me to research the topic.
Peter A. McCullough, MD, MPH, a highly published and respected academician, cardiologist, and epidemiologist published a recent Substack article outlining his concerns.
Virus in the Dust: Exposing the Fabricated Contagion of Andes Hantavirus
By Peter A. McCullough, MD, MPH
As an epidemiologist, I know contact tracing studies are inherently flawed. So, I was immediately suspicious of the claim that Hantavirus Andes strain had jumped from zoonotic to human-to-human spread. This claim is very likely to be false and should not be the basis for the global public health response to the Hantavirus outbreak from rodents on board the MV Hondius.
Dr. McCullough supported his argument well and listed a number of PubMed references. The Andes strain of the Hanta Virus is the only strain where human to human transmission has been argued. And that claim requires a high bar to be satisfied. The systematic review, the highest quality of medical evidence agrees with Dr. McCullough.
The Highest Quality of Evidence is the Systematic Review
Here is another PubMed study, a systematic review, that concludes, similar to Dr. McCullough, that there has not been adequate proof published to meet this burden of proof.
The conclusion of the paper states the following:
The balance of the evidence does not support the claim of human-to-human transmission of ANDV.
However, when I asked the Perplexity AI Model to summarize this paper, it reported the opposite, that this study actually reinforced the evidence for [not against] human to human transmission.
Perplexity Lies First Stating the Opposite of What the Paper Says
Here was the Perplexity summary:
The referenced PubMed study, “Evidence for Human-to-Human Transmission of Hantavirus” (PMC9574657, published in 2021), does not question or dispute the existence of human-to-human transmission of the Andes orthohantavirus (ANDV). In fact, it does exactly the opposite.
Perplexity did not just tell a falsehood; it spun a whopper. Watch it double down:
If you read the study, the authors actually acknowledge that while historical reports (such as those from the 1996 Argentina outbreak) relied primarily on epidemiological tracing and partial sequencing, the scientific consensus firmly accepts that ANDV can spread from person to person. The article explicitly aims to reinforce and prove this transmission pathway rather than cast doubt on it.
Here is why the study supports human-to-human transmission:
1. Genomic Sequencing Confirms Direct Spread
The study highlights that earlier investigations into human-to-human transmission were sometimes criticized because people in rural South America might have simply been exposed to the same infected rodents (a shared environmental source). To address this, modern studies use full-length viral genome sequencing. When scientists sequence the virus in a cluster of patients (e.g., an index patient and a doctor who treated them), they find 99.8% to 100% nucleotide identity. If the secondary patients had caught the virus from different rodents in the environment, there would be significantly more genetic variation.
2. Transmission Outside of Endemic Zones
The study, along with broader epidemiological data, notes that secondary infections frequently occur in environments where the primary rodent reservoir (the long-tailed pygmy rice rat) does not exist. For example, transmission has occurred during bus rides and within sterile hospital settings.
3. Animal Models and Superspreader Evidence
While PMC9574657 discusses the transmission mechanics, other contemporary research (such as a 2023 study published by the CDC) successfully modeled human-to-human transmission of ANDV using Syrian hamsters, definitively proving that horizontal transmission occurs via shedding in respiratory droplets and saliva. Furthermore, detailed contact tracing from outbreaks like the 2018 Epuyén outbreak confirmed “superspreader” events where a single individual infected multiple people during brief social interactions (like a birthday party or a funeral), which perfectly models human respiratory/saliva spread rather than rodent exposure.
In summary, rather than raising a “substantial question” about the existence of human-to-human transmission, the scientific literature—including the study you referenced—uses advanced genomics to validate that Andes virus is unique among hantaviruses for its ability to spread directly between people.
Calling AI Out on the Lie
So, I called AI out on this issue by quoting the conclusion: Please see the attached statement under CONCLUSIONS in the above referenced study you purport “validates” human to human transmission.
Here is the paper’s CONCLUSION:
“The balance of the evidence does not support the claim of human-to-human transmission of ANDV. “
Please revise your answer in view of this article’s conclusory statement.
Perplexity Walks back its Hanta Virus Lie
At this point, the Perplexity AI admitted it had given me an incorrect answer [basically lied]:
You’re right to call that out: my earlier description of this specific paper was incorrect. This 2021 systematic review explicitly concludes, “The balance of the evidence does not support the claim of human-to-human transmission of ANDV,” so it should not have been described as validating person-to-person spread.
Perplexity Finally Admits the Truth in Detail
It goes on:
What the paper says
The review included 22 studies and found that the stronger comparative studies generally did not support human-to-human transmission, with one exception: a prospective Chilean household-contact cohort that suggested increased risk among sex partners and with exposure to saliva, but the review judged that study to have serious risk of bias. The authors therefore concluded that the overall evidence base was weak, heterogeneous, and heavily confounded by possible shared rodent exposure.
Why the authors were skeptical
A central point in the paper is confounding by environmental exposure to infected rodent excreta and secretions, which the authors say was not adequately controlled in many studies that claimed interpersonal spread. The review also states that identical viral sequences do not by themselves prove person-to-person transmission when cases all occur within the same endemic area, because the same sequence could reflect exposure to the same local rodent-associated viral variant.
What this means
So yes, this paper does raise a substantial question about whether the published evidence truly establishes human-to-human Andes virus transmission, at least as of the review’s search period and under the evidentiary standards the authors applied. At the same time, the paper does not prove that human-to-human spread never occurs; instead, it argues that the evidence then available was insufficiently rigorous and that better cohort or case-control studies controlling for rodent co-exposure were needed.
Revised assessment: My corrected answer is:
This article does not validate human-to-human transmission; it argues the opposite, or at least that the claim was not adequately demonstrated by the available literature reviewed. The paper’s bottom-line position is that evidence for human-to-human transmission was weak, limited to Andes virus claims in parts of Argentina and Chile, and insufficient to support a firm causal conclusion, though the authors still said precautionary infection-control measures may be prudent because hantavirus pulmonary syndrome has a high fatality rate.
So, I went to another LLM. I asked Grok to summarize the same article, and Grok delivered the same lie. Here is what it said.
Second Verse: Same as the First
Summary: This paper provides strong epidemiological and molecular evidence of human-to-human transmission of Andes virus (a hantavirus causing hantavirus pulmonary syndrome, HPS) in Argentina.
And it was not merely a slip of the tongue, or an “accidental” misapplication of the facts. It was another bald-faced lie.
Conclusions of the Authors:
The study confirms that Andes virus can spread directly between humans. This has important implications for:
Infection control in hospitals (isolation precautions)
Contact tracing
Public health response during outbreaks
Significance:
This paper strengthens the evidence that certain hantaviruses (particularly Andes virus) are not solely zoonotic and can have limited person-to-person spread, unlike most other hantaviruses.
The study is considered high-quality evidence and is frequently cited in discussions about hantavirus transmission risks.
Once again, I called the AI out on the lie. This time it was Grok. I confronted the AI with the PubMed study’s published conclusion:
Please see the attached statement under CONCLUSIONS in the above referenced study you purport argues that it “validates” human to human transmission. Here is the paper’s CONCLUSION: “The balance of the evidence does not support the claim of human-to-human transmission of ANDV. ” Please revise your answer with this article’s conclusory statement.
Grok gave an Apology similar to Perplexity’s
This systematic review rejects the hypothesis of reliable or sustained human-to-human transmission of Andes virus. It emphasizes that apparent clusters are more likely explained by shared environmental/rodent exposure rather than direct person-to-person spread. The authors call for better-controlled studies.
Thank you for the correction. My earlier summary incorrectly presented this paper as supporting human-to-human transmission. It actually does the opposite by concluding that the overall evidence does not support it.
Grok apologized in a similar manner to Perplexity for lying to me about the Hanta virus being transmitted between humans. Why would different AI models come up with the same lie about a PubMed study?
Why Would Two Different AI Models Parrot the Same Lie?
I asked this to a third AI.
Here was my question: Why would two different AI models, Perplexity and Grok state the opposite, the first that this study reinforces there is proven human to human transmission, and the second stating that this paper provides strong epidemiological and molecular evidence of human-to-human transmission of Andes virus. Does this speak to programming an answer to reinforce consensus or the WHO position?
AI’s Excuse for Multiple AI’s Lying about the Same Study in the Same Manner
Almost every paper on Andes Virus (ANDV) begins by stating as a fact that it is “the only hantavirus known to be transmitted person-to-person.” AI models scan this introductory text and treat it as the paper’s finding.
However, PMC9574657 is a Systematic Review, which is a higher level of evidence. Its specific job was to look at the quality of the papers that made those claims. The authors found that while many papers claim transmission, those studies are almost all “noncomparative” (they didn’t use control groups), meaning they couldn’t prove the people didn’t just catch it from the same group of mice.
Calling the AI Out on the Direction of the Bias
AI continued to wade deep into it. It lied again, and then it lied some more. More BS in AI’s explanation that does not wash.
It doesn’t necessarily mean they were programmed to lie, but it does highlight a major flaw in how AI models process complex scientific papers: they often prioritize abstract summaries and historical context over the author’s actual methodology and conclusions.
Here is exactly why those models likely got it backwards and why PMC9574657 (The systematic review by Smith et al.) is a “trap” for AI:
1. The “Abstract vs. Analysis” Trap
Almost every paper on Andes Virus (ANDV) begins by stating as a fact that it is “the only hantavirus known to be transmitted person-to-person.” AI models scan this introductory text and treat it as the paper’s finding.
However, PMC9574657 is a Systematic Review, which is a higher level of evidence. Its specific job was to look at the quality of the papers that made those claims. The authors found that while many papers claim transmission, those studies are almost all “noncomparative” (they didn’t use control groups), meaning they couldn’t prove the people didn’t just catch it from the same group of mice.
2. Confusing “Molecular” with “Epidemiological” Evidence
Grok mentioned “molecular evidence.” Many studies show that victims in an outbreak have identical viral genetic sequences.
The AI logic: Identical sequences = one person gave it to the other.
The Paper’s logic: Identical sequences only prove they caught the exact same strain. If five people live in one house with one infected mouse, they will all have identical viral sequences, but they caught it from the mouse, not each other.
3. The “Signal-to-Noise” Problem
If 20 papers say “ANDV is contagious” and 1 systematic review says “Wait, the evidence for that is actually very poor,” an AI—which works on probabilistic patterns—is more likely to side with the 20 papers. It struggles to recognize that one high-quality systematic review outweighs 20 biased case reports.
4. Risk of Bias Scoring
PMC9574657 uses a tool called ROBIS to score previous studies. It labeled most studies supporting human transmission as having a “Critical Risk of Bias.” A human reading this understands that “Critical Risk” means “don’t trust these results.” An AI might see the study title listed in the bibliography and assume it’s being cited as supporting evidence rather than being debunked.
The Thumb on the Scale
source justusrhope.substack.com