Glyphosate: What the Evidence Says About the World's Most-Used Herbicide

Glyphosate is the most widely used herbicide in history — 8.6 billion kilograms applied globally through 2014, with 74% of that total sprayed in the preceding decade Benbrook 2016. Usage has continued rising since. It is detectable in the urine of 81% of Americans aged six and older — CDC NHANES 2013-2014 of Americans aged six and older Ospina et al. 2022. And it is the subject of the most consequential disagreement in modern chemical safety science: the IARCInternational Agency for Research on Cancer says it is probably carcinogenic. The US EPAEnvironmental Protection Agency said it wasn't — and a federal court threw that finding out. The article you're reading is part of the broader landscape of environmental chemical exposure. Glyphosate is the entry where the science and the politics are hardest to separate.

Glyphosate (C₃H₈NO₅P, CASChemical Abstracts Service 1071-83-6) is a broad-spectrum, non-selective herbicide — it kills nearly any plant it touches by inhibiting the EPSPS enzyme5-enolpyruvylshikimate-3-phosphate synthase — an enzyme in the shikimate pathway that plants and bacteria use to make essential amino acids. Humans lack this pathway, which is why glyphosate was initially considered harmless to animals. in the shikimate pathway. Plants use this pathway to make three essential amino acids. Block it, and the plant dies within days. Humans lack the shikimate pathway entirely, which is why glyphosate was originally considered harmless to animals — a clean kill for weeds with no mechanism to affect people. That argument has not aged well, but more on that below.

That framing held for decades. Monsanto patented glyphosate in 1974 and sold it as Roundup. In 1996, the company introduced Roundup Ready crops — genetically engineered soybeans, corn, cotton, and canola designed to survive glyphosate spraying, so farmers could kill weeds without killing the crop. Usage exploded. By 2014, global application had risen roughly 15× from 1974 levels. Then, in 2015, two of the world's leading scientific authorities looked at the same chemical and reached opposite conclusions.

One detail worth holding onto: glyphosate doesn't simply vanish after it has done its job. In soil and water it slowly breaks down into AMPAAminomethylphosphonic acid (CAS 1066-51-9) — the primary degradation product of glyphosate. AMPA persists in soil roughly three times longer than glyphosate itself, accumulates with repeated application, and is regulated under the same toxicological reference values., which persists roughly three times longer than the parent compound and is regulated by the EU under the same toxicological reference values rather than separately. When environmental scientists test rivers, soil, or human urine, they typically find both.

Almost every conversation about glyphosate makes the same quiet substitution. A regulator approves "glyphosate." A farmer sprays "Roundup." A toxicologist tests "the active ingredient." These are not the same thing. GlyphosateThe herbicidal active ingredient — a polar molecule that on its own struggles to penetrate the waxy cuticle of a plant is the active ingredient. Roundup is the active ingredient plus surfactants — wetting agents that help it cross the waxy cuticle of a leaf and the lipid bilayer of a cell. The surfactants are not inert. They have their own toxicology, and in many studies, they are doing more biological work than the chemical the regulator approved.

The historical surfactant in Roundup was POEAPolyethoxylated tallow amine — a non-ionic surfactant derived from animal fats, used for decades as the wetting agent in Roundup formulations. Restricted in EU glyphosate products in 2016 (Commission Implementing Regulation (EU) 2016/1313)., polyethoxylated tallow amine. In a study of nine commercial pesticide formulations, MesnageRobin Mesnage, molecular toxicologist at King's College London — has authored more than 30 peer-reviewed papers on pesticide formulation toxicity and colleagues found that 8 of 9 formulations were up to 1,000× more toxic than their active ingredients alone were up to 1000× more toxic to human cells than their declared active principles Mesnage et al. 2014. The active ingredient on the regulatory file is not what's in the bottle.

Two years later, Defarge and colleagues asked a more specific question: if you dose human placental cells with the formulation surfactants alone — no glyphosate at all — what happens to aromataseCytochrome P450 19A1 — the enzyme that converts androgens to oestrogens. Inhibition of aromatase suppresses oestrogen synthesis and is the mechanism behind several breast-cancer drugs., the enzyme that synthesises oestrogen? The answer was that aromatase activity dropped at concentrations 800× lower than agricultural-use dilutions — Defarge et al. 2016 on POEA and APG lower than agricultural-use dilutions Defarge et al. 2016. Both the original POEA and the alkyl polyglucoside (APG) replacement showed the effect. The endocrine signal lives in the surfactant, not the active ingredient that the regulator evaluates.

The EU restricted POEA in glyphosate products in 2016 Commission Implementing Regulation (EU) 2016/1313. Manufacturers reformulated. The standard playbook: substitute one less-tested surfactant for another, with no fresh long-term toxicology required. Banned in Europe. Replaced with surfactants that have been studied in less depth than the one they replaced. The same pattern that produced BPA-free plastics that leach BPS — a regulatory swap on the active ingredient that does not engage the toxicology of the supporting cast.

The primary consumer route isn't the garden. It's the kitchen. Pre-harvest desiccationThe practice of spraying a herbicide on a crop shortly before harvest to kill the plant and accelerate drying. Common on oats, wheat, barley, lentils, and chickpeas. Prohibited in the EU's 2023 glyphosate renewal but still permitted in the US, UK, and Canada. — spraying glyphosate on grain crops shortly before harvest to accelerate drying — accounts for only about 2% of total agricultural use but over 50% of dietary exposure. The spray hits the grain itself, days before it's harvested, milled, and turned into the cereal, bread, flour, and beer you consume.

Here's the part that surprises most people: the USDAUnited States Department of Agriculture does not routinely test for glyphosate in its annual Pesticide Data Program — despite it being the most widely applied agricultural chemical in the country. The testing data above comes from independent laboratories. The most-used pesticide in history is the one the government doesn't routinely check for in food.

In March 2015, a working group of 17 experts from 11 countries convened at IARCInternational Agency for Research on Cancer in Lyon and classified glyphosate as Group 2A — probably carcinogenic to humans. The classification rested on three pillars: limited evidence of cancer in humans (primarily non-Hodgkin lymphomaA group of cancers of the lymphatic system — the cancer type most consistently associated with glyphosate exposure in occupational epidemiology), sufficient evidence in experimental animals, and strong evidence of genotoxicity. The working group reviewed only published, peer-reviewed studies.

Five years later, the EPA issued its interim registration review decision: 'not likely to be carcinogenic to humans.' The EPA included unpublished, industry-commissioned regulatory studies in its assessment — studies submitted by the registrant (Monsanto) as part of the product approval process. Of the regulatory genotoxicity studies the EPA relied on, 99% were negative. Of the 118 peer-reviewed genotoxicity studies reviewed by IARC — 83 positive, 35 negative peer-reviewed genotoxicity studies IARC reviewed, 83 were positive — about 70% Benbrook 2019. Same chemical. Different libraries. Opposite conclusions.

Then the Ninth Circuit Court of Appeals weighed in. On 17 June 2022, in Natural Resources Defense Council v. EPA, the court vacated the human-health portion of EPA's interim decision — not because the court found glyphosate dangerous, but because the EPA had failed to follow its own guidelines and reach a conclusion supported by substantial evidence. The ecological portion was sent back without vacatur. As of 2026, no EPA cancer determination for glyphosate is legally in force. The registration review continues.

The epidemiological evidence centres on one cancer: non-Hodgkin lymphoma. Zhang and colleagues pooled six studies — including the large Agricultural Health Study cohort — and found a 41% increased risk of non-Hodgkin lymphoma in the highest-exposure subgroups across pooled studies — Zhang et al. 2019 increased risk of NHLNon-Hodgkin lymphoma — a group of cancers of the lymphatic system among the highest-exposure subgroups (meta-RR 1.41, 95% CI 1.13–1.75) Zhang et al. 2019. That headline figure — the one Bayer's lawyers have spent five years contesting in court — sits on top of a more complicated literature.

The AHSAgricultural Health Study — a US National Cancer Institute prospective cohort of licensed pesticide applicators in North Carolina and Iowa, enrolled 1993-1997 is the largest study of actual glyphosate users ever conducted. NCINational Cancer Institute researchers enrolled 54,251 licensed pesticide applicators across North Carolina and Iowa between 1993 and 1997, asked them in detail about pesticide use, then followed them for cancer incidence through 2012 in North Carolina and 2013 in Iowa — more than a decade of additional follow-up beyond the previous analysis. 44,932 of 54,251 applicators reported using glyphosate in the Agricultural Health Study cohort of them — 82.8% — had used glyphosate. By the end of follow-up, 5,779 of them had developed cancer.

Andreotti and colleagues published the result in 2018. For NHL, the cancer at the centre of the IARC classification, the highest-exposure quartile showed no statistically significant association — RR 0.87, 95% CI 0.64–1.20. The authors' bottom line was direct: 'no association was apparent between glyphosate and any solid tumours or lymphoid malignancies overall, including NHL and its subtypes.' But there was one signal that wouldn't go away. AMLAcute myeloid leukemia — a cancer of the bone marrow that produces abnormal white blood cells. In the highest exposure tertile at a 20-year exposure lag, AML risk was elevated: RR 2.04 highest exposure tertile, 20-year lag — Andreotti et al. 2018 Agricultural Health Study (95% CI 1.05–3.97, P-trend 0.04) Andreotti et al. 2018. The authors flagged it as 'requiring confirmation.' This is the strongest single piece of evidence pointing toward harm in the strongest study design — and it's not the cancer the lawsuits have been about.

Then there are the Monsanto Papers. In 2017, 141 internal Monsanto documents were released during Roundup cancer litigation. They revealed that Monsanto employees had drafted or heavily edited a landmark 2000 safety review — Williams, Kroes, and Munro, published in Regulatory Toxicology and Pharmacology — without being listed as authors. The paper concluded Roundup 'does not pose a health risk to humans.' It became one of the most-cited papers on glyphosate safety. Government public health agencies worldwide cited it without caveat. Its retraction was finally posted by Regulatory Toxicology and Pharmacology in November 2025 — twenty-five years after publication, eight years after the ghostwriting was made public, four months after Alexander Kaurov and Naomi Oreskes formally requested the journal act. The safety paper the world relied on was ghostwritten by the company selling the product.

The litigation built outward from there. In August 2018, a San Francisco jury awarded Dewayne 'Lee' Johnson, a school groundskeeper diagnosed with NHL after years of spraying Roundup on athletic fields, $289 million — later reduced on appeal to $20.5 million. In March 2019, federal jurors in Hardeman v. Monsanto — the first MDL bellwether — returned an $80 million verdict, eventually reduced to $25.3 million and affirmed by the Ninth Circuit in 2021. Two months after that, in Pilliod v. Monsanto, an Alameda County jury awarded a husband and wife $2.055 billion after both were diagnosed with NHL following decades of Roundup use; the trial court substantially reduced the punitive damages on post-trial motion and the case continued through California's First District Court of Appeal.

Bayer, which acquired Monsanto in 2018, has paid more than $10 billion settling Roundup cancer lawsuits to date. On 17 February 2026 the company announced a proposed nationwide class settlement of up to $7.25 billion, paid out over up to 21 years, to resolve current and future NHL claims tied to pre-February-2026 exposure. The Circuit Court of the City of St. Louis granted preliminary approval on 4 March 2026, with the final fairness hearing scheduled for 9 July 2026. The litigation is not over — opt-outs are still possible — but the global trajectory is the largest mass-tort settlement framework in modern US chemical-safety history.

The original safety case for glyphosate rested on a single biochemical fact: humans don't have the shikimate pathway, so the chemical that kills plants by blocking that pathway has nothing to bind to in our bodies. The argument was clean. It was also incomplete. The shikimate pathway is present in your gut bacteria — the gut microbiotaThe community of microorganisms living in the human digestive tract — roughly as numerous as human cells, contributing to immune function, vitamin synthesis, and metabolic regulation. Many gut bacteria use the shikimate pathway that glyphosate inhibits. that ferment your fibre, train your immune system, and produce roughly half of your serotonin. Glyphosate doesn't need to inhibit a human enzyme to do something to a human body. It just needs to inhibit the enzymes inside the bacteria a human body depends on.

Mesnage and colleagues tested this directly. They gave Sprague-Dawley rats either glyphosate alone or the commercial formulation Roundup MON 52276 in their drinking water at three doses — 0.5, 50, and 175 mg/kg body weight per day — for 90 days. After three months they sampled the cecum, the part of the rat's gut where bacterial fermentation peaks. The signature was unmistakable: shikimic acid and 3-dehydroshikimic acid had accumulated in the gut contents at the higher doses Mesnage et al. 2021. These two metabolites are what builds up immediately upstream of the EPSPS enzyme when EPSPS is inhibited — the same biochemical fingerprint glyphosate leaves behind in a dying weed, observed inside a mammal.

A 2023 study in mice asked whether the effect appears at doses humans actually encounter. Lehman and colleagues exposed C57BL/6J mice to glyphosate at concentrations from 1 to 100 µg/mL — overlapping the US acceptable daily intake — and reported gut microbiota composition shifts, increased pro-inflammatory CD4+IL17A+ T cells, and elevated lipocalin-2 (a marker of intestinal inflammation), alongside reduced Bifidobacterium pseudolongum and Lactobacillus Lehman et al. 2023. The shikimate pathway humans don't have. The gut bacteria that have it. The doses regulators consider safe.

What this work doesn't yet show is a downstream human disease endpoint. The rats that accumulated shikimic acid in their cecum did not develop colitis or NHL within 90 days, and rodent microbiome shifts do not translate cleanly into human ones. The honest reading is that the original safety argument — 'humans don't have the target enzyme' — is no longer the end of the conversation, but the next step (linking gut-microbiome disruption to a specific human outcome at realistic exposures) hasn't been taken yet. That gap is exactly where regulatory hazard assessment lives: a plausible mechanism, intermediate-marker evidence, and an unfinished bridge to clinical disease.

And the formulation work returns here. The same surfactants that disrupt aromatase in placental cells (Defarge 2016) are also the molecules that help glyphosate cross intestinal lipid barriers. Mesnage and Antoniou argued in 2017 that ignoring co-formulant toxicity systematically falsifies the safety profile of commercial pesticides — that the active-ingredient-only review the EPA conducts isn't testing the product the consumer is exposed to Mesnage & Antoniou 2017. The endocrine endpoint, the gut endpoint, the carcinogenicity endpoint — three separate research lines, three separate signals, all converging on the same point: the chemical the regulator approved and the product the public encounters are not the same thing.

The regulatory picture is fractured. Commission Implementing Regulation (EU) 2023/2660, adopted on 28 November 2023, renewed glyphosate authorization for ten years — through 15 December 2033 — and prohibited use as a desiccant. The maximum annual application rate was cut from 2.16 to 1.44 kg of active ingredient per hectare. The US has no cancer determination currently in force. The UK's Health and Safety Executive extended approval to 15 December 2026, with a major public consultation underway and a final decision expected in late 2026.

The precautionary position: the IARC classification is peer-reviewed and based on published evidence. The EPA's alternative conclusion has been vacated. The Andreotti AHS data is honest counter-evidence on NHL but flags an unresolved AML signal at long exposure lag. The mechanistic case has shifted in five years — from 'glyphosate has no human target' to 'glyphosate hits the shikimate pathway in your gut bacteria, and the formulation surfactants disrupt aromatase below agricultural-use levels.' If organic alternatives exist and are accessible, the precautionary cost of choosing them is low. This isn't a strong recommendation in the way BPA or lead are — the human cancer evidence is contested. But when plausible mechanisms exist on multiple fronts, the strongest cohort flags an unconfirmed signal, and the alternative costs the same per serving, the eso-friendly approach is to reduce where you can.

Glyphosate is the chemical where this library can't give you a clean answer. The evidence is real but contested. The regulatory system is fractured. The largest cohort study found no NHL association but flagged an AML signal at long exposure lag. The shikimate pathway humans were said to lack turned out to be the same one their gut bacteria depend on. The company that sold the product ghostwrote its own safety review — and it took a quarter of a century to retract.

The practical response is proportionate: choose organic for the foods where desiccation residues are highest — oats, wheat, chickpeas, lentils. The cost difference per serving is small. The chemicals that aren't contested — the ones where the evidence is decades deep and the editorial position is clear — are the ones to act on first. Glyphosate is the one to watch.