A February 18, 2026, Executive Order (EO) from President Trump, entitled “Promoting the National Defense by Ensuring an Adequate Supply of Elemental Phosphorus and Glyphosate-Based Herbicides,” concerns phosphorus’ use as a critical mineral in fertilizers for crop production (and the manufacture of explosives, among other things), and the role phosphorus plays in the production of the herbicide glyphosate, also used in crop production. Glyphosate is indeed important due to the way the United States currently produces many crops — some of the most widely grown crops (corn, soybeans, and cotton) now have 90 percent of their acreage planted with herbicide-resistant seeds, predominantly glyphosate, to help control weed pests.

Phosphorus and its important role in food production as a nutrient is not surprising from the view that such critical materials should be available through domestic production; the United States currently relies on imports for a large portion of these needs. The Fertilizer Institute estimates that about 35 percent of the U.S. supply of phosphate fertilizer has been imported over the last five years. It also reports that China and Russia produce 52 percent of global phosphate fertilizers — providing the potential for domestic or global trade disruption via defense security issues. The White House also issued a fact sheet alongside the EO that reiterates the need to ensure an adequate supply of herbicides. What caught observers by surprise was the inclusion of glyphosate — front and center in the EO — as a defense security issue.

To research the question of how vulnerable domestic production of glyphosate might be, we used an artificial intelligence (AI) tool (e.g., Gemini) to ask about how glyphosate is made (two main different production methods) and about supply, import, and other key factors. The answers from the AI tool were straightforward initially and included references from government agencies, manufacturers, or suppliers of the needed materials (see this link to the AI query). Along the way, the AI tool “asked” if the designation as a critical material could affect the liability for damages resulting from production of the designated material — specifically: “Would you like to know how these liability protections specifically intersect with ongoing Roundup-related litigation?”

The response goes on to tie the designation under the Defense Production Act as a liability shield for future production of such designated materials. It then includes some discussion of the current legal liabilities for the current manufacturer of glyphosate, Bayer Chemical. Litigation about the safety of glyphosate and court cases against Bayer are widely reported, and the litigation has already cost the company billions of dollars with more expected to come.

This blog piece is not about the safety of glyphosate per se, but of note is that the White House chose to issue this EO at a time when the Make America Healthy Again (MAHA) community has been especially critical of the Administration, even when still considered an important political base for the President. This EO comes after the “MAHA Moms” have already been upset over the final MAHA report issued last September. For more information, see our August 28, 2025, blog item. A number of environmental non-governmental organizations (eNGO) have long disputed the U.S. Environmental Protection Agency’s (EPA) cancer assessment of glyphosate and have new-found allies within the MAHA movement.

Lessons from and about AI-Generated Statements

Given the controversy over the safety of glyphosate, especially regarding its cancer assessment conclusions and any liability shield, the AI query then asked: “Would you like to know how the recent retraction of a 25-year-old safety study might impact this 2026 EPA re-evaluation?” The answer includes that “The 2026 re-evaluation of glyphosate has been complicated by the recent retraction of a 1999 safety study originally used by the EPA to support its ‘non-carcinogenic’ finding. This retraction impacts the EPA’s ongoing Interim Registration Review…. [T]he retracted study, which focused on the genotoxicity of glyphosate, was a cornerstone of the EPA’s position that the chemical does not damage DNA.”

This seemed a curious statement from the AI string — that it was a “cornerstone” of EPA assessments — given the very many studies that EPA has reviewed, as part of its assessments, approximately 15,000 submitted studies according to EPA guidelines (docket EPA-HQ-OPP-2009-0361-0086) along with volumes of outside literature. The EPA press secretary, meanwhile, was quoted in The Washington Post as saying: “The retraction of this publication has no impact on EPA’s glyphosate assessment, which has historically reviewed more than 6,000 individual studies across all disciplines, including human and environmental health.” (6,000 is, to our ears, an under-estimate.)

EPA has had a focus on the results of epidemiological studies and other important components of its assessment of the carcinogenic risk of glyphosate, and held a three-day peer review session of its Scientific Advisory Panel (SAP) in December 2016 to review its evaluation of the carcinogenic potential of glyphosate.

The later-retracted study was part of that assessment reviewed at the SAP meeting, and the EPA document notes that the study in question, among others, “…[was] funded and/or linked to Monsanto Co. or other registrants.” (The issue of unreported conflicts of interest with Monsanto at the time of publication appears to be the basis or a basis for the scientific journal’s retraction decision.)

What was perhaps most striking about the AI-generated statement that this study was a “cornerstone” of the current EPA position is that the word “cornerstone” appears in The New York Times article on the matter. Since AI pulls information based on sources that receive the most hits and not necessarily the ones that are the most accurate, search results in a space that utilizes AI can be misleading. It is a reminder that AI-generated statements that appear as straightforward conclusions are often, in reality, more akin to a popularity contest amongst the most common or frequent comments or conclusions found across (mostly Internet) sources.

This is an example of the number of citations that repeat the description of the retracted study on glyphosate as “important,” “sweeping,” “bedrock, or “cornerstone” (all terms in various press articles on the matter) almost certainly having a direct impact on EPA’s conclusions. It is also a reflection of the issue’s popularity far more than it reflects a change in the underlying science.