Glyphosate: the "Weight of Evidence" does not buy the right to stay Silent

On 8 September 2026, ECHA's Committee for Risk Assessment ruled on a new rat study from the Ramazzini Institute's Global Glyphosate Study. Its verdict: the study doesn't change what RAC already found in May 2022 — glyphosate still doesn't warrant a carcinogen classification. That's the third time in under a decade the same hazard classification has survived a fresh challenge. The interesting part isn't that the answer held. It's what had to happen before anyone could say so.

The review didn't start with a study. It started with a letter that went unanswered. In October 2023, Ramazzini publicised early results from its multi-arm Global Glyphosate Study. ECHA and EFSA asked for the underlying data. They got nothing — not then, and not after a follow-up in February 2024.

That changed once the carcinogenicity results were formally published, as Panzacchi et al., Environmental Health 24, 36 (2025), on 10 June 2025. Two weeks later the European Commission stepped in: a joint mandate, issued by DG SANTE's Claire Bury on 24 June, gave ECHA 15 months to update its 2022 opinion once it had the data, and gave EFSA six months after that to issue its own statement. Both agencies accepted within three days.

What they asked Ramazzini for next, on 8 July, is worth reading closely: the full study report with individual-animal data; complete details on the substance tested, including its origin; and — flagged as critical — the exact product, purchase date and EU authorisation number, since the study used a specific commercial formulation, Roundup Bioflow. That last point matters. The EU approves glyphosate as an active substance; Member States approve the actual products built on it. So any findings tied specifically to Roundup Bioflow only count once its identity is confirmed — that doesn't touch the study's separate findings on pure glyphosate, but for the formulation-specific results, regulators first needed to know exactly what had been tested.

Ramazzini's reply, on 19 July, is the essay's real hinge. There is no separate study report — the Institute said so plainly. Only the published paper and its raw data exist, because the carcinogenicity arm is the one finished piece of a project whose other two arms are still running; a full report won't exist until they finish too. Ramazzini promised the raw data by 30 September. It arrived ten days late, on 10 October. Even then, what came was a peer-reviewed paper and a large set of data tables — not the detailed study report that lets an outside assessor rebuild an experiment from scratch. That gap doesn't disqualify the study: weight-of-evidence reviews can and do draw on published, non-GLP literature. But a study report is still the normal anchor document of regulatory toxicology, and going without one makes a study harder to check.

That gap probably explains more than any dispute over the science itself. Classification doesn't ask whether one study shows harm. It asks whether the totality of evidence — weighed for reliability, relevance and consistency — still points the same way. Commission Implementing Regulation (EU) 2023/2660 grounds glyphosate's approval in "two comprehensive assessments since 2012": a decade of accumulated, cross-checked material. A single new study has to move that balance in combination with everything else — and this one arrived with an open question about whether it was even testing the product regulators use. RAC hasn't published its reasoning yet. This account describes the evidence it had going in, not how it weighed it. But the evidence alone is telling.

Zoom out, and the 2023 renewal tells a similar story at a larger scale. Approval nearly doubled in length, from the five truncated years granted in 2017 to ten years, running to December 2033. The rapporteurship changed too: instead of one Member State assessing the file, a group of four — France, Hungary, the Netherlands and Sweden — did it jointly, an option created in 2019 and used here, for the first time, on a file this contested. And the renewal quietly pushed a long list of decisions down to national level: co-formulants, residues in rotational crops, groundwater protection, exposure for small mammals, spray drift — and, for effects on biodiversity through the food chain, an outright admission that no EU-level method exists yet. Until one does, each Member State decides for itself.

Marta Morvillo, writing in 2020, argued that the glyphosate dispute hadn't really changed the EU's approach to regulatory science — the underlying risk-analysis model held, and reforms reached only the "sources" and "levels" of how evidence gets made, never the harder question of what counts as acceptable protection. She said a full test of that claim would have to wait for the next renewal. It's now happened twice: once in the 2023 architecture, once in the 2025–26 Ramazzini review. Both times, RAC's hazard classification held. For 2025–26, that's still only half the picture — RAC's reasoning isn't published, and EFSA's own statement isn't due until six months after it is. What did move, both times, was how much of the actual judgment the EU handed elsewhere: to a joint rapporteurship, to national authorities, to a data-request process that took three months just to get the raw data on the table before assessment could even start.

Both sides, in fact, keep making the same mistake. ECHA announced its 2026 conclusion on 8 September, but the opinion behind it — the actual reasoning — won't be published for, in the agency's own words, "a couple of months" more. That gap is exactly the space press releases and competing narratives rush to fill, not because anyone is necessarily wrong, but because nobody outside the committee can check the reasoning yet. It's precisely the failure the Commission's 2002 guidelines on the use of expertise were meant to prevent. Ramazzini did the same thing first, publicising its findings in October 2023 before giving the agencies anything to evaluate them against. Neither shortcut is a failure of science. Both are failures of sequence — verdict before argument — and each one hands the other side, and the press, exactly the excuse needed to claim the result was decided in advance.

What's actually left open isn't glyphosate's classification. It's whether this cycle — mandate, letters, months of data exchange, reconfirmation — is now the EU's default answer to any single-lab challenge. It's whether either side will break the habit of announcing conclusions before publishing the reasoning behind them. And it's whether the biodiversity-methodology gap the 2023 renewal left open gets closed before 2033, or simply gets handed to Member States like everything else on that list. None of that is a small cost: asking citizens to trust a verdict nobody outside the committee can yet check is exactly how public understanding of science erodes.

How can scientific institutions stay trusted while their own communication becomes increasingly performative?

Three fixes would do more than any statement of principle. First, publish reasoning and verdict together — or don't announce until you can. Second, hold every challenger, institute and agency alike, to the same evidentiary template: an OECD-style study report, not just a peer-reviewed paper. Third, register the test material up front, publicly, before results come out — the product, its batch, its provenance — so "what was actually tested" is never a question anyone has to chase down after the fact.

Sources: ECHA, "Glyphosate" hot-topics page (accessed 10 September 2026) — echa.europa.eu/hot-topics/glyphosate- · Commission Implementing Regulation (EU) 2023/2660 of 28 November 2023 — eur-lex.europa.eu · Panzacchi, S., Tibaldi, E., De Angelis, L. et al., "Carcinogenic effects of long-term exposure from prenatal life to glyphosate and glyphosate-based herbicides in Sprague–Dawley rats," Environmental Health 24, 36 (2025) — doi.org/10.1186/s12940-025-01187-2 · Morvillo, M., "Glyphosate Effect: Has the Glyphosate Controversy Affected the EU's Regulatory Epistemology?" European Journal of Risk Regulation 11 (2020) 422–435 — doi.org/10.1017/err.2020.11 · European Commission–ECHA–EFSA mandate and correspondence, EFSA-Q-2025-00410 (24 June – 10 October 2025) — open.efsa.europa.eu/questions/EFSA-Q-2025-00410