How flawed animal models undermine glyphosate risk assessments — and why New Zealand regulators may be relying on evidence that was compromised before it was even published.
When the Environmental Protection Authority or the Ministry for Primary Industries tells New Zealanders that glyphosate is safe at current exposure levels, they are not conducting their own independent experiments. They are, by and large, reviewing and deferring to a body of existing scientific literature — much of it produced by or funded by the agrochemical industry — and accepting the preclinical animal studies within it as a valid basis for setting human exposure limits.
But what if the animal models those studies rely on are not neutral scientific instruments? What if the very creatures used to establish “safe” doses are themselves so biologically distorted by the conditions of their laboratory existence that their responses to toxic substances bear only a passing resemblance to what happens inside a human body?
That is not a fringe question. It is increasingly being raised by toxicologists, epidemiologists, and biomedical researchers who are confronting what some describe as a structural crisis at the heart of preclinical science. And it has direct, urgent implications for every New Zealander who eats food grown with glyphosate-based herbicides, drinks water from catchments where those herbicides are used, or lives near land that is routinely sprayed.
“This article draws substantially on the analysis of Dr Ronald Kostoff, whose study The Credibility of Animal Models for Human Toxicology Studies provides a comprehensive multi-species examination of these distortions.”
How Glyphosate Safety Limits Are Set — and What That Process Assumes
The standard regulatory pathway for determining whether a chemical is safe for human exposure runs through the animal laboratory. A substance is administered to rodents — typically rats or mice — at varying dose levels over weeks, months, or years. Researchers observe what happens. Tumors, organ damage, reproductive effects, neurological changes: all are noted. A “no observed adverse effect level” is then identified, an Uncertainty Factor is applied to account for the gap between rats and humans, and a maximum acceptable daily intake is declared.
This process is presented as rigorous, objective, and scientifically grounded. MPI and the EPA cite it constantly. Regulators in Brussels, Washington, and Canberra cite it too. It forms the bedrock of the reassurances we are given about glyphosate.
Here is what that process quietly assumes: that the laboratory rat or mouse used in these studies is a reasonable biological stand-in for a human being. Similar enough in its physiology, its metabolism, its hormonal systems, and its cellular biology, that what poisons the rat will likely threaten the human, and what spares the rat will likely spare the human too.
That assumption deserves far more scrutiny than it receives.
Why Laboratory Mice Are Not Reliable Models for Human Glyphosate Exposure
The rats and mice that populate the world’s toxicology laboratories are not simply small humans. Nor are they even typical members of their own species. They are, in a meaningful biological sense, artificial creatures — shaped by decades of commercial breeding practices that have inadvertently but profoundly altered their physiology in ways that directly affect how they respond to toxic substances.
Consider telomere length. In wild mice, telomeres — the protective caps on chromosomes that regulate cellular ageing and replication — operate under natural evolutionary constraints. Shorter telomeres act as a biological brake on runaway cell division, which is one of the body’s key defences against cancer formation. Longer telomeres permit more cell division and tissue repair, but at the cost of greater cancer risk over a lifetime.
Commercial breeding programs have, entirely accidentally, disrupted this balance. By continuously breeding only young, reproductively active animals and culling them before they age, breeders have applied sustained selective pressure favouring longer telomeres and higher reproductive output. Over generations, this has produced laboratory mice with dramatically elongated telomeres compared to their wild counterparts — essentially hyper-regenerative animals whose cells divide with unusual frequency and whose tissues repair themselves at rates that no human cell could match.
What does this mean for toxicology? It cuts in two directions, both of them distorting. First, because these mice are unusually capable of repairing cellular damage, compounds that cause serious tissue injury in human organs may appear harmless in the laboratory — the mice simply regenerate what was damaged before the researchers can measure the harm. Second, because these mice are disproportionately prone to tumour formation regardless of what they are exposed to, some compounds may be flagged as carcinogenic in rodent models that would not trigger the same effect in human tissue with normally functioning telomeric biology.
In either case, the data generated is systematically skewed away from human reality. And this is before we even consider the other ways laboratory conditions distort the animals being studied.
How Lab Conditions — Stress, Artificial Light, and Confinement — Distort Toxicology Results
Standard laboratory rats — primarily the Sprague-Dawley and Wistar strains used in the majority of toxicology studies — are raised under artificial 12-hour light/dark cycles using fluorescent lighting. Because rats are nocturnal, this chronic light exposure suppresses pineal gland melatonin production. That might sound like a sleep problem. It is actually a metabolic catastrophe.
Melatonin is not simply a sleep hormone. It is a primary antioxidant, a regulator of insulin sensitivity, and a synchroniser of the body’s peripheral organ clocks — the biological timekeepers in the liver, kidneys, adrenal glands, and gut that coordinate when metabolic processes run and when they rest. When melatonin is chronically suppressed, these peripheral systems fall out of synchrony with each other. The liver no longer metabolises substances on the same schedule as the immune system. The gut microbiome — itself a primary site of xenobiotic metabolism, meaning the place where foreign chemicals including pesticide residues are first transformed — shifts toward inflammatory, dysbiotic communities that process drugs and toxicants differently than a healthy gut ecosystem would.
The practical consequence for glyphosate research is significant. When a rat’s liver drug-metabolising enzymes are operating on a desynchronised, circadian-disrupted schedule, the absorption, distribution, metabolism, and excretion profile it generates for any chemical — including glyphosate — may bear essentially no resemblance to what happens in the rhythmically coherent physiology of a human being eating breakfast, absorbing residues from their cereal, and processing those residues through an intact and normally functioning gut and liver.
Then there is the cage itself. Standard laboratory housing — small, barren, plastic boxes — denies animals any opportunity for the natural behaviours that regulate their stress hormones: digging, climbing, exploring, running, hiding, and social interaction. The result is chronically elevated corticosterone — the rodent equivalent of cortisol — which restructures the animal’s physiology at every level. It suppresses immune function, alters gut barrier integrity, promotes visceral fat accumulation, and changes how the liver processes foreign substances. A rat in a standard laboratory cage is not a healthy baseline animal. It is an animal in a state of chronic physiological stress, and its responses to toxic exposures reflect that stressed baseline, not normal mammalian biology.
Glyphosate Is Never Tested the Way Humans Are Actually Exposed to It
Even if we set aside the biological distortions in the animals themselves, there is a second fundamental problem with how safety limits for glyphosate are derived: they are calculated from experiments that expose animals to glyphosate alone, in isolation from everything else.
Real New Zealanders are not exposed to glyphosate alone. They are exposed to glyphosate formulations — which include surfactants and adjuvants that are often more acutely toxic than the active ingredient — alongside residues from other agricultural chemicals, microplastics, heavy metals, air pollutants, food additives, and the cumulative effects of medications, alcohol, and environmental stressors. These co-exposures do not simply add up. They interact. They synergise. Combinations of toxicants at doses that appear safe in isolation can cause profound physiological damage when they are experienced together, because they compete for the same detoxification pathways, overwhelm the same organs simultaneously, or amplify each other’s mechanisms of harm.
Single-stressor laboratory studies cannot capture this. They are not designed to. And the Uncertainty Factors applied by regulators to bridge the gap between animal data and human populations do not account for it either. The standard Uncertainty Factor — typically set somewhere between 10 and 1,000 depending on the data available — is supposed to compensate for the difference between species and for variation within the human population. It says nothing about synergistic toxicity. It says nothing about cumulative lifetime exposure. It says nothing about the pregnant woman, the child, the elderly person with compromised liver function, or anyone whose toxic burden at the time of exposure is already elevated.
What this means is that even a perfectly conducted single-stressor glyphosate study, using the most biologically realistic animals in the best-equipped laboratory, would still generate a safety threshold that is structurally unable to protect people in the real world.
The Ramazzini Glyphosate Study: Cancer Signals Regulators Are Dismissing
Here is what makes the current regulatory situation in New Zealand particularly difficult to defend: even within the limitations of the conventional animal model framework, concerning findings about glyphosate keep emerging — and regulators keep finding reasons to look away.
The Ramazzini Institute in Italy, a respected independent toxicology organisation with no industry funding, recently published a two-year rat study exposing animals to glyphosate at the EU’s current Acceptable Daily Intake — the dose regulators have declared safe. The results included a range of tumours and, perhaps most strikingly, early-onset leukemia, with roughly 40 percent of leukemia cases occurring in animals that were less than one year old. In human developmental terms, that corresponds to young adulthood.
The predictable response from industry and some regulators was that “rats are not humans.” And as we have explored above, that is literally true in ways more profound than those critics appreciate. But that same logic, applied consistently, should also undermine confidence in the animal studies that established the safety limits in the first place. You cannot selectively invoke the limitations of animal models to dismiss studies that raise concerns while simultaneously accepting animal studies as definitive proof of safety. Either the models are valid enough to tell us something meaningful about human risk, or they are not. Regulators cannot have it both ways.
The scale of the replication problem makes this harder to ignore. In a landmark analysis, scientists at Amgen found they could not reproduce 89 percent of highly cited preclinical cancer studies — and Bayer, examining a broader range including cardiovascular and metabolic research, found a 75 percent failure rate. The replication crisis is not confined to oncology; it reflects systemic problems in how preclinical animal research is designed, analysed, and reported across biomedical science. These are not minor statistical wobbles. They are the foundational outputs that regulatory safety assessments — including those used to declare glyphosate safe — treat as authoritative.
The Monsanto Papers: How Glyphosate Safety Data Was Controlled and Manipulated
The biological and methodological problems described above are serious enough on their own terms. But there is a further dimension that New Zealand regulators appear equally reluctant to confront: the question of who controls the raw data that safety assessments are built upon, and what incentives they have to present it accurately.
This is not a hypothetical concern in the context of glyphosate. Court proceedings in the United States — specifically the wave of litigation brought by people who developed non-Hodgkin lymphoma after occupational glyphosate exposure — produced internal Monsanto documents that revealed a pattern of conduct that goes well beyond ordinary commercial self-interest. These documents showed evidence of ghostwriting scientific papers that were then published under the names of independent academics, of direct attempts to influence regulatory reviewers, and of internal awareness of safety concerns that were not reflected in the company’s public communications or regulatory submissions.
This matters for animal model research specifically because the choice of which studies to submit to regulators, which results to highlight, and how to frame findings from ambiguous experiments is entirely at the discretion of the registrant. A corporation with billions of dollars riding on a favourable safety assessment has both the motive and, historically, the demonstrated willingness to select animal models, housing conditions, exposure protocols, and statistical approaches that are most likely to produce the regulatory outcome it needs. If a particular rat strain is less sensitive to glyphosate-induced liver damage, that strain gets used. If a particular study duration is too short to capture long-term carcinogenic effects, that duration becomes standard. None of this requires outright fabrication. It requires only the systematic application of methodological choices that, individually, appear defensible, but collectively tilt the evidentiary playing field.
What makes this especially problematic for New Zealand is that MPI and the EPA do not have independent access to the raw underlying data from the proprietary studies submitted as part of product registration. They review summaries and reports prepared by or for the registrant. When they defer to EFSA or the US EPA, they are deferring to bodies that have themselves been the subject of serious and documented criticism regarding their handling of industry-submitted glyphosate data — including specific allegations that key studies showing harm were dismissed using methodological criteria that were not applied with equivalent rigour to studies showing safety.
The question this raises is uncomfortable but necessary: if the animal studies underpinning glyphosate’s approved status were selected, designed, and submitted by an organisation with a demonstrated history of managing the scientific record in its commercial interest, and if those studies used the biologically distorted laboratory animals and single-stressor methodologies described throughout this article, what exactly is the independent scientific foundation on which New Zealand’s regulators are standing?
What NZ’s MPI and EPA Are Actually Relying On for Glyphosate Safety
When New Zealand’s regulators assess glyphosate, they lean heavily on reviews conducted by overseas bodies — the European Food Safety Authority, the US EPA, the Joint FAO/WHO Meeting on Pesticide Residues. These reviews, in turn, synthesise large bodies of literature, much of it produced by or submitted by Monsanto and its successor Bayer as part of the original product registration process.
This raises a question that New Zealand regulators have not publicly and satisfactorily answered: to what extent are the core animal studies underpinning New Zealand’s glyphosate safety assessments based on standard laboratory strains, housed in standard barren cages, exposed to single stressors in isolation, assessed using Uncertainty Factors that ignore synergistic real-world exposures?
The answer, almost certainly, is: almost entirely. That is what “standard regulatory toxicology” means. The distortions described in this article are not edge cases or theoretical concerns. They are features of the mainstream preclinical methodology that produced the data New Zealand’s safety assessments are built upon.
What would it look like for MPI or the EPA to acknowledge this? It would mean admitting that the foundational evidence base for glyphosate safety is systematically biased in ways that may underestimate harm — arising from both the structural limitations of preclinical methodology and, as the Monsanto litigation documents suggest, from the way that methodology has been selectively deployed by those with a financial interest in the outcome. It would mean looking seriously at the growing literature on real-world multi-stressor interactions. It would mean demanding independent access to raw study data rather than accepting registrant-prepared summaries as a substitute for scientific transparency.
There is no sign that either agency is moving in this direction.
Is Glyphosate Safe in New Zealand? The Questions Regulators Have Not Answered
We are not arguing here that glyphosate is definitively dangerous at the levels most New Zealanders are currently exposed to. That would be overstating what the science currently shows. What we are arguing is something more fundamental and more unsettling: that the scientific framework used to tell us it is safe is built on models that are known to be compromised, assessed through methodologies that are known to be inadequate for real-world exposure scenarios, and overseen by regulatory bodies that have not publicly grappled with these limitations.
The Ramazzini findings, the mounting epidemiological literature linking glyphosate exposure to non-Hodgkin lymphoma, the studies connecting glyphosate to gut microbiome disruption in ways that matter for human immune function — none of these prove that current New Zealand exposure levels are causing harm. But they raise questions that deserve rigorous, independent, and transparent investigation.
What if the animal studies that underpin glyphosate’s safety approval were conducted using animals whose biology was already significantly distorted from normal mammalian physiology? What if the synergistic effects of glyphosate combined with other chemicals New Zealanders routinely encounter have never been properly studied? What if the Uncertainty Factors applied by regulators are arithmetically inadequate to bridge the gap between a stressed, circadian-disrupted laboratory rat and a pregnant woman in Hawke’s Bay eating produce from a glyphosate-sprayed orchard?
These are not rhetorical questions. They are scientific questions that have not been answered. And until they are, the confidence with which New Zealand’s regulators declare glyphosate safe deserves to be treated not as settled science, but as an institutional position in urgent need of scrutiny.
This article draws on peer-reviewed toxicology literature and published critiques of preclinical research methodology. We encourage readers to consult the primary sources and form their own conclusions.
Further Reading
On animal model limitations and the replication crisis
- Begley, C.G. & Ellis, L.M. (2012). Raise standards for preclinical cancer research. Nature, 483, 531–533. nature.com/articles/483531a — The landmark Amgen paper documenting that only 11% of 53 “landmark” cancer studies could be reproduced.
- Kostoff, R.N. (2026). The Credibility of Animal Models for Human Toxicology Studies: an AI-based Update. TrialSite News. trialsitenews.com — Comprehensive multi-species analysis of how laboratory breeding and housing conditions systematically distort preclinical biology.
- Weinstein, B.S. & Ciszek, D. (2002). The reserve-capacity hypothesis: evolutionary origins and modern implications of the trade-off between tumour-suppression and tissue-repair. Ageing Research Reviews, 1(3), 615–627. — On the telomere trade-off and why laboratory mouse cancer biology diverges from human biology.
- OncoBites (2020). Modeling Aging and Cancer: Are Lab Mice different From Their Wild Cousins? oncobites.blog — Accessible summary of the elongated telomere problem in standard laboratory mouse strains.
On the Ramazzini Global Glyphosate Study
- Panzacchi et al. (2025). Leukemia in Sprague-Dawley Rats Exposed Long-term from Prenatal Life to Glyphosate and Glyphosate-Based Herbicides. Environmental Health. — The peer-reviewed publication of the full carcinogenicity findings.
- Global Glyphosate Study press release on leukemia data (2023). glyphosatestudy.org
- George Mason University (2025). International Study Reveals Glyphosate Weed Killers Cause Multiple Types of Cancer. gmu.edu
- Environmental Health News (2025). Glyphosate exposure at low doses linked to leukemia and multiple tumors in new rat study. ehn.org
On the Monsanto Papers and corporate manipulation of science
- US Right to Know: The Monsanto Papers. usrtk.org/monsanto-papers — Archive of internal documents released during litigation, including ghostwriting evidence.
- Science (2025). Journal retracts weed killer study backed by Monsanto, citing ‘serious ethical concerns‘. science.org — The retraction of the 2000 paper ghostwritten by Monsanto scientists.
- Glenna et al. (2021). Suborning science for profit: Monsanto, glyphosate, and private science research misconduct. Research Policy. sciencedirect.com
- Mind the Gap (2020). Case Study: Monsanto ghost-writing and funding scientific research. mindthegap.ngo
On glyphosate and the gut microbiome
NMGNZ (2025). Glyphosate and the Gut Microbiome: What the Science Says. nomoreglyphosate.nz
Walsh, L., Hill, C. & Ross, R.P. (2023). Impact of glyphosate (Roundup™) on the composition and functionality of the gut microbiome. Gut Microbes. ncbi.nlm.nih.gov/pmc/articles/PMC10561581
RSC Food & Function (2024). Effects of glyphosate exposure on intestinal microbiota, metabolism and microstructure: a systematic review. pubs.rsc.org
Image Source & Attribution
We’re grateful to the talented photographers and designers whose work enhances our content. The feature image on this page is by marisha5.
