Industries / Mining and materials / Chemicals and petrochemicals / Fertilizers and agrochemicals
Crop protection chemicals
Crop protection chemicals are the herbicides, insecticides and fungicides that kill weeds, insects and plant diseases; this work is finding new ones, proving to regulators that they are safe enough to sell, and making them.
7
Milestones achieved, from 1910 to 2025.
2
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1910
A law against weak poisons
Paris green and lead arsenate were the two most used insecticides of the early 1900s, and without purity standards some brands were too weak to be of any value while others were strong enough to injure the crops they were meant to protect. Six states had passed purity laws when the US Insecticide Act, signed on 26 April 1910, made it a misdemeanor to make or ship adulterated or misbranded insecticides and fungicides, with the US Department of Agriculture collecting and testing samples. Manufacturers backed the law, partly because one bad product hurt the whole market. It policed purity, not safety to people or wildlife. In 1972 the Federal Environmental Pesticide Control Act required federal registration of every pesticide product and made the plants that make pesticides register and report how much they made and sold.
LessonThe first pesticide law protected buyers from fraud; rules on safety for people and wildlife came decades later.
Credit
the US Congress, with support from manufacturers, growers and entomologists
1973
Sunlight-stable insecticides
Pyrethroids are man-made relatives of the insecticides found in pyrethrum flowers. In November 1973 Michael Elliott's team at Rothamsted, England, published "A Photostable Pyrethroid" in Nature. The team went on to develop bioresmethrin, permethrin, cypermethrin and deltamethrin. In the 1980s these made up about two-thirds of the world pyrethroid market, and pyrethroids held over 25% of the insecticide market, according to John Pickett's Royal Society memoir of Elliott.
Credit
Michael Elliott and colleagues at Rothamsted
1980s
Weedkillers by the gram
DuPont sold the first sulfonylurea herbicide, chlorsulfuron, as Glean for wheat in the early 1980s. It blocks ALS, an enzyme that plants need to make three amino acids and that animals do not have. It was first tested at several ounces per acre, but field trials showed it worked at much lower rates, according to an Ohio State University course text.
LessonA very potent product with one target can wear out fast; a University of Nebraska lesson says DuPont withdrew Glean from crop use five years after release because resistant kochia, a weed, spread.
Credit
DuPont, now part of Corteva Agriscience
1996
Fungicides modeled on nature
The strobilurins, a class of fungicides modeled on a group of natural compounds that kill fungi, were first launched in 1996. By 2002 the class included azoxystrobin, the world's biggest-selling fungicide, and six active ingredients were on sale, according to a review by Syngenta scientists.
Credit
Syngenta, which made azoxystrobin
2004
Consent before export
The Rotterdam Convention came into force on 24 February 2004, after its adoption in 1998. It covers pesticides and industrial chemicals that member countries have banned or severely restricted. Once a chemical is listed, each importing country gets a guidance document and has nine months to say whether it will accept imports, and exporting countries must make their exporters follow those decisions. FAO and UNEP had run a voluntary version, called prior informed consent, since 1989.
Credit
FAO, UNEP and the parties to the convention
2018
Neonicotinoids banned outdoors
Neonicotinoids, insecticides that act on insects' nicotine receptors, were the fastest-growing insecticide class in the 2000s, by a 2008 review from Bayer scientists. The EU restricted three of them, clothianidin, imidacloprid and thiamethoxam, in 2013 to protect honeybees. After the European Food Safety Authority (EFSA) confirmed the risks in February 2018, the EU banned all their outdoor uses from 30 May 2018, leaving only permanent greenhouses.
LessonAn approval granted on the evidence of its day can lose most of its uses when new evidence about other species arrives.
Credit
the European Commission and EU member states, on EFSA's assessments
2025
A new herbicide group
In April 2025 Peru approved FMC's Keenali herbicide, made with a new active ingredient, tetflupyrolimet. It blocks DHODH, an enzyme plants need to make some of the building blocks of DNA. The Herbicide Resistance Action Committee (HRAC), an industry body that sorts herbicides by how they work, lists it as the only member of Group 28 on its 2026 chart. FMC called it the first new herbicide mode of action in three decades. It is not registered in the US: in February 2026 EPA asked for public comment on requests from Arkansas and Missouri to use it on up to 646,000 acres of rice under emergency exemptions, against barnyardgrass resistant to several herbicides.
Credit
FMC
Watch for
Icafolin's claimed new action
Bayer says icafolin-methyl is "agriculture's first new mode of action herbicide for post-emergent weed control for broadacre crops in over 30 years". It has applied for approval in the US, Canada, Brazil and the EU, and plans to launch first in Brazil in 2028. In its December 2025 update HRAC placed icafolin-methyl in an existing group, Group 23, which it renamed "Microtubule Interference - Unclear Site of Action" because the exact target is not known.
Achieved whena regulator approves it and researchers outside Bayer publish the site it acts on, or HRAC gives it a group of its own.
Claimed by
Bayer, August 2025
AI finds new targets
Enko Chem says its ENKOMPASS platform, which combines DNA-encoded libraries (huge mixtures of tagged test molecules), machine learning and structural biology, has found new modes of action for fungicides and insecticides, "up to 75% faster and at 90% lower cost" than usual research. In December 2025 Enko said a fungicide class found with the platform and developed with Syngenta would go to field tests at several sites the following year, based on greenhouse and lab data. No outside check of the speed or cost figures is cited.
Achieved whenSyngenta or an outside party publishes field results for the fungicide, or a molecule from the platform is filed for registration with its target confirmed by others.
Claimed by
Enko Chem, October 2026; Enko and Syngenta, December 2025
Open
01
Mode-of-action test set
A public benchmark that asks a model to predict a pesticide's mode of action, the plant, insect or fungal process it blocks, from its chemical structure. It would hold every active ingredient classified by the industry's resistance committees (HRAC for herbicides, IRAC for insecticides, FRAC for fungicides), each with its structure, and test models on ingredients classified after their training data, such as the five new herbicides HRAC added in December 2025.
The briefCredit · unclaimed02
Why EU approvals end
A table of every active substance that the EU has refused to approve or renew under its current pesticide law, Regulation (EC) No 1107/2009. Each row gives the reasons stated in the Commission's decision, coded into a short list (for example cancer or hormone hazard, risk to bees, groundwater, missing data), with links to the regulation and the EFSA conclusion behind it.
The briefCredit · unclaimed
Open 01 · the brief
Mode-of-action test set
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public benchmark that asks a model to predict a pesticide's mode of action, the plant, insect or fungal process it blocks, from its chemical structure. It would hold every active ingredient classified by the industry's resistance committees (HRAC for herbicides, IRAC for insecticides, FRAC for fungicides), each with its structure, and test models on ingredients classified after their training data, such as the five new herbicides HRAC added in December 2025.
Why it is still open
AI discovery firms publish speed claims, such as Enko's "up to 75% faster", that nobody can compare. Oršolić and colleagues (Scientific Reports, 2021) built random-forest models that sort herbicides by mode of action, for herbicides only and without a test on ingredients classified later. A 2023 review by Corteva scientists says AI-driven agrochemical discovery appears "to a far lesser extent in the publicly available agrochemical research literature compared to drug discovery". The nearest public benchmark, ApisTox (Adamczyk and colleagues, Scientific Data, January 2025), scores toxicity to honey bees, not mode of action. Searches of Europe PMC, arXiv and GitHub in November 2026 found no benchmark that spans herbicides, insecticides and fungicides or holds out newly classified ingredients.
Built with what exists
HRAC's herbicide master sheet and the IRAC and FRAC classification lists; PubChem's free API for structures; RDKit; Chemprop, an open model for molecular properties; a random forest like Oršolić and colleagues' as a baseline; an open-weight general model such as Mistral Large 4 as a comparison (see the Note "A trillion open parameters").
First step
Download HRAC's master sheet and match each herbicide name to a PubChem structure, listing the ones that fail.
Done looks like
Every classified active ingredient matched to a checked structure (HRAC alone reviewed 266 herbicide ingredients in 2024–2025), a held-out set of at least 30 ingredients classified since 2020, and published accuracy for at least three baseline models, with the code.
Open 02 · the brief
Why EU approvals end
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A table of every active substance that the EU has refused to approve or renew under its current pesticide law, Regulation (EC) No 1107/2009. Each row gives the reasons stated in the Commission's decision, coded into a short list (for example cancer or hormone hazard, risk to bees, groundwater, missing data), with links to the regulation and the EFSA conclusion behind it.
Why it is still open
The EU Pesticides Database says whether each substance is approved and offers machine-to-machine APIs, but the reasons sit in the text of separate regulations and EFSA reports. AGRINFO, a service that tracks EU food and farm rules for low- and middle-income countries, summarizes recent non-renewals in reports that list the regulations, not in a coded table. Registration costs a new active ingredient US$41.8 million, 13.9% of the US$301 million total, in a 2024 survey of five large makers by AgbioInvestor for CropLife International. A Europe PMC search in November 2026 found no coded table of reasons.
Built with what exists
The EU Pesticides Database APIs; EUR-Lex for regulation texts; the EFSA Journal; a model built for extraction from long documents, such as Muse Ember (see the Note "The jobs agents repeat"), run by an agent that cites the sentence behind each code (see the Note "Findings that cite notes").
First step
Pull the list of non-approved substances from the database API and code the ten most recent decisions by hand to settle the coding scheme.
Done looks like
Every refusal and non-renewal coded, each code linked to the sentence it came from, a hand check of one row in five with no more than 2% wrong, and a published count of reasons by year.