The Reviewer

When the courthouse door closes, the regulator becomes the only institution left standing between the public and the product. So who reviewed the reviewer?

A COMPANION TO THE SHIELD | THE OLD GUARDIAN INVESTIGATION

The Shield documented what EPA’s glyphosate determination did to 100,000 plaintiffs. This piece examines how that determination was made, why a federal court found it defective, and what happened when nobody fixed it.

This piece is a companion to The Shield, a four-part Old Guardian series examining the legal, regulatory, and corporate architecture surrounding Roundup litigation. The Shield is available in full at theoldguardian.ca. Familiarity with the series is useful but not required. This piece stands on its own.

Fourteen Words

On June 25, 2026, the United States Supreme Court closed the courthouse door on more than 100,000 plaintiffs who alleged that exposure to Roundup caused their cancer. The ruling was seven to two. It was written by Justice Brett Kavanaugh. And it rested, in its entirety, on fourteen words.

The EPA has determined that glyphosate is not likely to be carcinogenic to humans.

Because the EPA reached that conclusion, the agency never required a cancer warning on Roundup’s label. Because no cancer warning was required, Monsanto had no federal obligation to include one. Because Monsanto had no federal obligation to include one, state tort claims demanding it are preempted by federal law. The logic is a chain. Remove any link and the chain breaks. The fourteen words are the first link. Everything that followed depended on them holding.

This piece is about those fourteen words. Where they came from. How they were produced. What a federal appeals court found when it examined the methodology behind them. And what the EPA did when that court sent them back for correction.

Before going further, one thing needs to be stated plainly. Monsanto’s record in this dispute is not a matter of serious debate. Internal documents revealed coordinated strategies to influence research and public messaging. Ghostwriting. Conflicts of interest. A landmark study retracted due to ethical concerns and reliance on unpublished Monsanto data. Juries across the country, hearing the evidence directly, found for plaintiffs repeatedly. Billions of dollars paid in settlements. That record stands and is not in dispute here.

But this piece is not about Monsanto. It is about the other institution in the room. The one whose job was to stand between Monsanto’s conduct and the public. The one with the authority, the mandate, and the methodology to reach a conclusion independent of what Monsanto wanted that conclusion to be. The EPA is not a passive actor in this story. It is the regulator. And its record, examined closely, raises questions that the Supreme Court ruling made more urgent, not less.

How the Process Works

Under the Federal Insecticide, Fungicide, and Rodenticide Act, every pesticide sold in the United States must be registered with the EPA. Registration requires the manufacturer to demonstrate that the pesticide, when used as directed, will not cause unreasonable adverse effects on human health or the environment. The EPA evaluates that demonstration and either grants or denies registration.

The evidentiary foundation of that evaluation is primarily built from studies submitted by the manufacturer. The company seeking registration bears the cost of producing the safety data the agency reviews. This is not a hidden arrangement. It is the explicit structure of the regulatory system. The EPA does not independently commission its own toxicology studies. It reviews what industry provides, supplemented by published scientific literature.

The obvious structural tension in that arrangement -- that the party with the most to gain from a favourable conclusion is also the primary supplier of the evidence the agency uses to reach it -- has been noted by scientists, regulators, and courts for decades. The scientists who gathered in Seattle in March 2026 to review the glyphosate evidence proposed an overhaul requiring data from independent laboratories free from conflicts of interest, with the costs borne by industry but the research conducted outside its control. That proposal has not been adopted.

Registration is not permanent. The EPA conducts periodic registration reviews to assess whether pesticides already on the market continue to meet current safety standards. For glyphosate, that review process has been running, in various forms, for decades. The relevant modern chapter begins with the EPA’s 2017 and 2019 reviews, which reaffirmed the not likely to be carcinogenic conclusion, and the 2020 interim registration decision that a federal court later vacated.

Built into the registration review process is a check: the Scientific Advisory Panel. The SAP is an independent body of external scientific experts convened to review EPA’s methodology and conclusions. It can identify problems, flag inconsistencies, and recommend corrections. What it cannot do is compel the agency to act on its recommendations. The SAP’s findings are advisory. The EPA is not required to follow them. In the glyphosate review, the SAP identified specific methodological problems with EPA’s approach. The agency proceeded anyway.

What the Cancer Paper Said

At the centre of EPA’s glyphosate cancer assessment is a document known internally as the Cancer Paper. It is the agency’s systematic review of available evidence on glyphosate’s potential carcinogenicity, produced as part of the registration review process. It reviewed epidemiological studies, animal carcinogenicity data, and mechanistic evidence. It arrived at a bottom-line hazard descriptor: not likely to be carcinogenic to humans.

What the Cancer Paper said on the way to that conclusion is the problem.

The epidemiological data the paper reviewed did not point clearly toward safety. Effect estimates across the case-control studies and meta-analyses the agency examined were greater than 1, meaning most studies indicated glyphosate exposure was associated with an increased risk of non-Hodgkin’s lymphoma. The meta-analyses showed effect estimates ranging from 1.3 to 1.5, indicating people exposed to glyphosate were between 30 and 50 percent more likely to develop NHL than those who were not. The paper reviewed that signal and attributed it to chance, bias, or confounding factors.

The specificity of that signal matters scientifically. Non-Hodgkin’s lymphoma is not cancer generally. It is a defined category of more than 60 distinct lymphoma subtypes affecting the lymphatic system. A chemical showing a consistent epidemiological signal for a specific cancer type is a more focused and harder-to-dismiss finding than one correlating with cancer broadly. Confounding factors and lifestyle variables are harder to invoke against a specific biological target. The NHL signal is not a vague association. It is a documented pattern across multiple independent research groups pointing at the same disease.

That is a legitimate scientific judgment. Discounting effect estimates for methodological reasons is standard practice in epidemiological review. What is not standard is what the Cancer Paper said next.

The paper explicitly stated that a conclusion regarding the association between glyphosate exposure and risk of NHL cannot be determined based on the available evidence. Those are the agency’s own words. The evidence was, by EPA’s own assessment, insufficient to reach a conclusion on the central question.

The EPA then selected not likely to be carcinogenic to humans as its final hazard descriptor.

Under EPA’s own Cancer Guidelines, that descriptor is appropriate only when the available data are considered robust for deciding that there is no basis for human hazard concern. The agency said in one paragraph that it could not determine the NHL risk. It said in effect in the next that the data were robust enough to conclude there was no meaningful concern. Those two statements cannot both be true. The Cancer Paper contained them both, and the EPA signed off on the document.

What the Court Found

In June 2022, the Ninth Circuit Court of Appeals examined EPA’s glyphosate cancer determination in Natural Resources Defense Council v. EPA, 38 F.4th 34. The court’s findings were not a matter of contested scientific interpretation. They were a documented accounting of specific methodological failures, each traceable to EPA’s own guidelines and its own Scientific Advisory Panel’s contemporaneous objections.

The first failure was the one already visible in the Cancer Paper. The court found EPA could not reasonably treat its inability to reach a conclusion about NHL risk as consistent with a conclusion that glyphosate is not likely to cause cancer. The internal contradiction was not subtle. It was the difference between “we cannot determine the risk” and “there is no meaningful risk.” The agency published both in the same document.

The second failure concerned animal carcinogenicity data. The EPA used historical control data selectively, invoking it only to undermine tumor results, never to support them. EPA’s own Scientific Advisory Panel had flagged this practice during the review. The panel warned that to subjectively choose to use historical control incidence data only in situations where it undermines tumor results is to potentially introduce bias into the assessment. The agency was told this by its own advisory body. It proceeded anyway.

The third failure is the most striking because it involved something the EPA appears to have invented. The agency discarded tumor results occurring at or above what it called a 1,000 milligrams per kilogram per day limit dose, treating those results as uninformative. The Ninth Circuit found that EPA’s own Health Effects Test Guidelines establish no such limit. The guidelines state only that the highest tested dose need not exceed that figure. They do not say results above it should be discarded. EPA created a threshold that does not exist in its own methodology and used it to eliminate data that pointed toward cancer risk. The Scientific Advisory Panel called it an ad hoc decision that is not well-justified and is not justified on the basis of the Cancer Guidelines.

The court’s conclusion was precise. Inconsistent reasoning is, absent explanation, the hallmark of arbitrary action. EPA’s glyphosate cancer determination did not survive substantial evidence review. The Ninth Circuit vacated the relevant portion of EPA’s 2020 interim registration decision and remanded it for a methodologically sound explanation.

Note what the court did not say. It did not say glyphosate causes cancer. It did not say EPA’s ultimate conclusion was wrong. It said EPA’s reasoning was internally contradictory, that it relied on data selectively in ways its own panel flagged as biased, and that it invented a methodological threshold to discard unfavourable results. It sent the determination back for a do-over conducted honestly.

What EPA Did About It

The Ninth Circuit issued its ruling in June 2022. EPA withdrew its interim registration decision in response. The withdrawal notice, filed by the agency itself, stated: EPA’s underlying scientific findings regarding glyphosate, including its finding that glyphosate is not likely to be carcinogenic to humans, remain the same.

That sentence deserves to sit alone for a moment. A federal appeals court found EPA’s reasoning internally contradictory, its data use selectively biased, and its invented limit dose methodologically unjustifiable. EPA’s response, in the document withdrawing the vacated decision, was to announce that its underlying finding was not changing.

Not: we are reviewing the specific methodology the court identified as defective. Not: we are examining whether the limit dose threshold we applied is supported by our own guidelines. Not: we are reconsidering how we used historical control data after our own Scientific Advisory Panel warned us against it. The finding remains the same.

Four years passed. As of 2026, EPA had not finalized a revised, methodologically corrected cancer determination. Glyphosate products remained on the market under existing labels. The registration review the agency described as anticipated to be completed in 2026 was a timeline that had been anticipated and missed before.

In March 2026, with Supreme Court oral arguments weeks away, EPA made a public commitment. Responding to questions from The New Lede following a scientific symposium at which CDC, National Cancer Institute, and university researchers called for urgent regulatory action, the agency stated: “This year, EPA will undertake a comprehensive, transparent, and rigorous scientific review of glyphosate to evaluate its use and ensure decisions are fully aligned with the best available science as well as human health and environmental protections. That review will include expanded public access to underlying data and science, analysis of independent peer-reviewed research on potential cancer and developmental risks, evaluation of opportunities to reduce exposure, and coordination with USDA, HHS, and states.”

Three months later, the Supreme Court ruled seven to two in Bayer’s favour, using EPA’s existing not likely to be carcinogenic determination as the controlling federal science. The promised rigorous review had not been completed. The vacated methodology had not been corrected. The finding a federal court found legally unsupportable was treated by the Supreme Court as dispositive.

And now the legal pressure that might have compelled the review to actually happen has been eliminated by the ruling the unrevised determination made possible.

The Structural Question

The Ninth Circuit identified not one methodological failure but three, each documented, each flagged in advance by EPA’s own Scientific Advisory Panel, and each working in the same direction. Every failure discounted data pointing toward cancer risk. None of the failures discounted data pointing away from it.

When a regulatory process produces a consistent pattern of errors that consistently benefit one party, the question that follows is not simply whether mistakes were made. It is why the mistakes always pointed the same way.

This piece will not answer that question definitively. The primary documents do not support a definitive answer, and this series does not make claims the evidence does not support. What the primary documents do support is a precise accounting of the structural conditions under which those directional errors became possible.

The evidentiary foundation of EPA’s review was primarily built from industry-submitted studies. The manufacturer with the most to gain from a favourable conclusion was the primary supplier of the data the agency reviewed. The independent check on that process, the Scientific Advisory Panel, issued specific objections to specific methodological choices. Those objections were advisory and non-binding. The agency was not required to address them. It did not.

The registration review timeline has no enforcement mechanism. When EPA misses anticipated completion dates, there is no consequence. The agency anticipated completing the glyphosate review multiple times before 2026 and did not complete it. Glyphosate products remained registered and on the market throughout. The manufacturer faced no regulatory consequence from the delay. Plaintiffs attempting to litigate in the interim faced a SCOTUS ruling that used the incomplete, court-vacated review as its scientific foundation.

The manufacturer’s own statements confirm the directional pattern the structural conditions made possible. In July 2021, Bayer CEO Werner Baumann told investors that more than 90 percent of Roundup litigation claims had come from the residential lawn and garden market. His explanation for removing glyphosate from consumer products was unambiguous: “This decision is exclusively geared at managing litigation risk and not because of any safety concerns.” The residential user -- the casual applicator who trusted the label, skipped the PPE, did not shower immediately after use, and wore work clothes the next day -- generated nine of every ten cancer claims. The manufacturer’s response was not to fix the label. It was to exit the market segment.

One further structural assumption embedded in the regulatory model deserves naming. The “safe when used as directed” conclusion was built on studies using fresh, properly formulated product applied under controlled conditions. Glyphosate products degrade over time in storage. The surfactants and adjuvants that drive herbicidal efficacy -- the speed at which a product kills weeds visibly -- break down with age, temperature fluctuation, and exposure. But herbicidal degradation and carcinogenic potential are two separate biological mechanisms. The glyphosate molecule itself is chemically stable, non-volatile, and does not degrade in the presence of light or air. A bottle that has lost weed-killing speed has not lost glyphosate concentration. The consumer who uses more of an older product to compensate for reduced visible efficacy may be increasing their exposure to the compound of concern, not decreasing it. No label communicates this distinction. No regulatory model accounts for it.

The Tracey Woodruff observation, made at the March 2026 Seattle symposium, is worth returning to here. Woodruff is a professor at Stanford University and a former senior scientist for EPA’s own Office of Policy. She said: “It’s hard to get the government to change, particularly because the industry spends a lot of money lobbying Congress to make sure the laws favor their point of view.” That is not an activist claim. That is a former EPA scientist describing from the inside the structural dynamic the outside record confirms.

The structural conditions that allowed three directional methodological failures to survive EPA’s internal review process, survive the Scientific Advisory Panel’s contemporaneous objections, survive a federal court vacatur, and ultimately survive intact into a Supreme Court ruling are not accidental features of the system. They are the system. Industry-submitted evidence. Non-binding independent review. No enforcement timeline. No consequence for delay. And a legal architecture, now confirmed by the Supreme Court, that treats the output of that system as the controlling ceiling on what any American court can find.

What Comes Next, and a Note on Canada

The EPA committed in March 2026 to a comprehensive, transparent, and rigorous scientific review of glyphosate. That commitment is now the primary accountability lever remaining after the Supreme Court ruling eliminated the legal pressure to deliver it. Whether the review happens, whether it honestly addresses the three specific methodological failures the Ninth Circuit identified, and whether its conclusions are reached through independent evidence rather than industry-submitted data are the live questions this investigation will continue to monitor.

Congressional oversight of EPA’s registration review process is a second lever. The specific failures documented here -- the invented limit dose, the selective historical-control use, the internal contradiction between inconclusiveness and the not likely descriptor -- are on the public record in a published federal appellate decision. They are subjects that fall within congressional oversight authority. Whether any committee pursues them is a political question. The record exists regardless.

A note on Canada. EPA’s glyphosate determination does not exist in isolation. Canada’s Pest Management Regulatory Agency conducts its own registration reviews and has historically arrived at conclusions that mirror EPA’s on glyphosate, including its own not likely to be carcinogenic finding. That parallel raises a question the PMRA has not been required to answer publicly: did Canada’s review reach its conclusion independently, through its own methodological assessment of the available evidence, or did it proceed substantially from EPA’s prior determination? If Canada’s regulatory independence on glyphosate is procedural rather than substantive, the structural questions this piece raises about EPA apply with equal force north of the border. That question deserves its own investigation, and The Old Guardian intends to pursue it.

For now, the American record stands on its own. An agency whose methodology a federal court found internally contradictory and legally unsupportable. A withdrawn decision whose underlying finding the agency immediately announced was not changing. Four years without a corrected determination. A Supreme Court ruling built on the uncorrected foundation. And a promise of rigorous review made three months before the ruling that eliminated the pressure to keep it.

The fourteen words that closed the courthouse door on 100,000 plaintiffs came from somewhere. This piece has followed them back to their source. What was found there is not reassuring. And the institution responsible for those words has, so far, given no indication it intends to revisit them.

Editor’s Notes

Companion series: This piece is a companion to The Shield, a four-part Old Guardian series. Part One: “From Trust the Science to Shield Us from the Lawsuits” (published prior to the Durnell ruling). Part Two: “Seven to Two” (the ruling and the dissent). Part Three: “Trust the Regulators” (the EPA determination and the Ninth Circuit findings). Part Four: “The Architecture” (the full containment map). All four parts are available at theoldguardian.ca.

Primary sources: Natural Resources Defense Council v. U.S. Environmental Protection Agency, 38 F.4th 34 (9th Cir. 2022) (full opinion); Monsanto Co. v. Durnell, U.S. Supreme Court, decided June 25, 2026 (majority opinion by Kavanaugh, J.); EPA withdrawal notice following Ninth Circuit vacatur, 2022; EPA public statement on glyphosate registration review, March 2026, as reported by The New Lede (Carey Gillam, March 27, 2026) and verified against the original statement; EPA Cancer Paper, glyphosate registration review (internal language quoted via Ninth Circuit opinion at 38 F.4th 34); Scientific Advisory Panel findings on glyphosate methodology (cited within the Ninth Circuit opinion); International Agency for Research on Cancer, glyphosate classification as probable human carcinogen, 2015; Tracey Woodruff, Stanford University and former EPA Office of Policy senior scientist, statement at Seattle Glyphosate Symposium, March 2026, as reported by The New Lede; Bayer CEO Werner Baumann, investor call statement, July 29, 2021 (as reported by C&EN, AgWeb, and Modern Farmer); Bayer AG, Five-Point Plan to Close the Roundup Litigation, May 2021.

Confidence tier: All findings attributed to the Ninth Circuit are drawn from the published decision at 38 F.4th 34. EPA’s withdrawal notice language is drawn from agency records as cited in court filings. The EPA March 2026 statement is quoted verbatim as verified against The New Lede’s March 27, 2026 report. The Baumann investor call quote is drawn from contemporaneous reporting across multiple outlets and from Bayer’s own published Five-Point Plan documentation. The herbicidal degradation versus carcinogenic potential distinction reflects established chemical stability data for glyphosate and is not an assertion about human health outcomes. No claim in this piece asserts definitively that glyphosate causes cancer in humans. The Canada/PMRA paragraph identifies a question for future investigation and does not assert a conclusion.

Ongoing monitoring: The Old Guardian will continue to monitor the EPA glyphosate registration review, the PMRA parallel review question, paraquat and chlorpyrifos preemption proceedings, and the $7.25 billion Roundup settlement fairness proceedings.

The Old Guardian | theoldguardian.ca