An MIT-led research team is set to receive up to $53 million from the federal Advanced Research Projects Agency for Health to build next-generation alternatives to glyphosate and other widely used weed killers. The project, called Precision Learning for On-target Weed control, or PLOW, will lean on generative AI and engineered microbes to try to outpace the chemistry that has dominated American farms for decades.
The funding comes through ARPA-H's STREAM program, according to MIT News, and as LAVX News notes, STREAM stands for Systems for Tracking and Resilience in Efficient Agricultural-input Management. PLOW is expected to draw researchers from MIT, the Broad Institute of MIT and Harvard, the University of North Carolina at Chapel Hill, the University of Tennessee at Knoxville, the University of Illinois at Urbana-Champaign, and Robigo. Christopher Voigt, head of MIT's Department of Biological Engineering and the project's principal investigator, said herbicide resistance mutations and supply-chain risks could threaten the agrochemicals farmers currently rely on, per the same MIT account.
Why Farmers Keep Reaching for the Same Chemistry
More than 1 billion pounds of herbicide are applied across the United States every year, the MIT report states, and those chemicals are credited with preventing large crop losses from weeds while supporting regenerative farming practices that protect soil. Glyphosate, the most familiar of these herbicides, kills weeds by blocking EPSP synthase, an enzyme plants need to manufacture several critical amino acids.
But that reliance has a cost. Some weeds have acquired genetic mutations that let them survive repeated herbicide use, a trend the MIT article attributes directly to the chemicals' own widespread application. A U.S. Department of Agriculture Economic Research Service chart adds that farmers in every state surveyed under USDA's 2019 Agricultural Resource Management Survey reported declining glyphosate effectiveness, though that data point carries lower confidence in the available record. Separately, a research summary indexed on PMC notes that two-thirds of all the glyphosate sprayed in the U.S. between 1974 and 2014 was applied in just the final decade of that span, underscoring how quickly use has accelerated.
Health Exposure and a Phosphorus Supply Problem
MIT News reports concerns that herbicide exposure may be associated with cancer, metabolic disorders, and endocrine disorders, with farm workers facing frequent contact and members of the public potentially encountering residues through food and drinking water. These are reported concerns, not proof that exposure causes those conditions. For glyphosate specifically, the EPA says its assessment of risk under expected real-world exposures differs in scope and method from IARC’s assessment of whether a chemical can cause cancer under any conditions. Voigt also cited phosphorus dependence as a supply-chain vulnerability for glyphosate production, with much of that phosphorus imported.
The U.S. Environmental Protection Agency requires certain toxicity tests before a pesticide product can be registered.
ARPA-H wrote on X that “what protects crops shouldn't put people at risk,” framing STREAM as a joint effort with HHS, the EPA, and USDA to cut herbicide use, track exposure, and remove residues faster.
How the AI and Microbial Pieces Fit Together
PLOW’s proposed weed-control candidates would begin as computational predictions, not ready-to-use products. According to MIT News, a team led by Regina Barzilay, Connor Coley, and Alexander Tropsha will use generative AI to identify chemical compounds or small interfering RNA molecules that could disrupt plant enzymes targeted by existing herbicides. The models could produce billions of candidates, which the team plans to prioritize for resistance durability, low-cost synthesis, and reduced risks to people and ecosystems.
Those predictions will be tested and refined in stages. MIT News reports that a screening team led by Angela Koehler and Jaime Cheah will look for molecules that actually hit their intended targets, with the results feeding back into the models. Greenhouse work led by Neal Stewart and Scott Lenaghan, alongside Patrick Tranel, Tim Willson, and Andee Wallace, is intended to provide an early test in plants; the team expects its first greenhouse tests of initial RNAi candidates to conclude by the end of the program’s first year.
Microbes Designed to Clean Up After Agrochemicals
A separate part of PLOW would use engineered bacteria and fungi, as well as small RNA, to break down agrochemicals and formulation ingredients. MIT News describes the intended outcome as conversion of leftover materials into biomass, a proposed way to reduce residues reaching food, soil, or waterways. The project’s aims are not evidence that these systems have yet proved effective or safe.
What Would It Take to Reach Farms?
Greenhouse trials would not by themselves authorize commercial use. The EPA’s registration manual says data requirements vary according to the pesticide category being reviewed, so the specific requirements for PLOW’s proposed chemical or microbial products cannot be determined from the project description alone. The agency also describes experimental-use permits as a separate process requiring EPA approval, distinct from authorization for commercial pesticide use.
There is a U.S. precedent for registering an RNA-interference pesticide, but it is not a test of PLOW’s approach to weeds. The EPA announced in December 2023 that it had registered Ledprona, an RNA-interference biopesticide targeting Colorado potato beetles on potato crops. That insecticide decision does not establish the effectiveness or safety of PLOW’s proposed herbicide candidates.
Per the MIT News disclosure, ARPA-H is funding the work under agreement number 140D042690030, and the announcement's authors noted that the views and conclusions expressed do not represent official U.S. government policy. The available evidence does not establish that PLOW's AI-generated candidates or microbial systems have been proven effective, safe, or ready for regulatory approval — the greenhouse results due within the program's first year will offer the first real test of whether the approach can move beyond the lab.
