Small Farms Worry Their Data Could Be Used Against Them, So a Case Western Team Is Building an AI Sandbox That Keeps It Private

Farm technology runs on data: how often a field is watered, which fertilizer went down and when, what a piece of equipment cost. That information can help an AI model suggest better practices. It can also, in the wrong hands, tell a supplier how much a farmer is willing to pay or tell a buyer how badly a farmer needs to sell.

A cybersecurity researcher at Case Western Reserve University in Cleveland is trying to separate the two. Erman Ayday, an associate professor in the Department of Computer and Data Sciences at the Case School of Engineering, has received a four-year grant of nearly $600,000 from the U.S. Department of Agriculture’s National Institute of Food and Agriculture to build a more secure AI framework for small and mid-sized farms, the university announced on Oct. 20.

“Digital agriculture is the future of sustainable food production, but if the technology only benefits the largest players, we risk severely damaging the diversity of our agricultural ecosystem,” Ayday said in the announcement. “No farmer should have to choose between modernizing their farm and compromising their privacy.”

The worry behind the project

According to the university, small- and mid-sized operations make up more than 90 percent of U.S. farms, and many of them have been hesitant to adopt AI tools out of concern that the data those tools collect could be misused. The announcement lists the kinds of information farmers consider sensitive: what equipment they use and what they paid for it, how often they water, which fertilizers and pesticides they apply and when they buy them. Farmers, it says, worry that such details could lead to price discrimination, higher insurance rates or a competitive disadvantage.

Large corporate farms can afford their own data security. Smaller family operations generally cannot, and the university says that gap is part of what drew Ayday to the problem. He was working on research for ICICLE, a National Science Foundation–funded AI institute focused on intelligent cyberinfrastructure, when he noticed that most digital agriculture research centered on crop yield and largely ignored how to protect the data farm programs collect.

“We recognized that the agricultural sector completely lacked the kind of robust, federated data protection frameworks we see in healthcare or finance,” Ayday said.

The university cites a global survey released in September that found 23 percent of North American farmers reported using AI to help inform their day-to-day operations. The announcement does not name the survey, so we could not check its method.

How the sandbox is supposed to work

Ayday’s co-investigator is Alfonso Morales, the Vilas Distinguished Achievement Professor in the Department of Planning and Landscape Architecture at the University of Wisconsin–Madison, who studies food systems. The university says the two will consult directly with farmers and lead a team building a digital sandbox, a closed-off computing environment where farm data can be studied without being exposed.

The project builds on work the team has already published. In a paper in the proceedings of the 2026 Conference on Computer Science and Intelligence Systems, Osama Zafar, Rosemarie Santa González, Mina Namazi, Morales and Ayday described a privacy framework that combines two techniques. Dimensionality reduction compresses a farm’s data into a smaller set of features, and differential privacy adds carefully calibrated random noise so that no single farm’s details can be pulled back out. The authors wrote that they tested the approach on real-world datasets and found it held up against adversarial attacks while performing comparably to a system that pooled all the raw data in one place.

A second paper, posted to the research repository arXiv by Zafar, Santa González, Morales and Ayday, describes the working platform, called the Digital Agriculture Sandbox. Its design is easier to picture than the math. Each farmer uploads data to a private account as a spreadsheet-style CSV file, a format the authors chose because many sensors and farm management systems already export it. Other farmers cannot see it.

The platform can then do a few things without moving that raw data around. It can show a farmer a ranked list of other farmers whose data looks similar, identified only by username, and let them chat if they choose. It can train a shared machine learning model through federated learning, a method in which each farm’s data trains a small local model and only the model updates, not the data, are combined. And each model in the system’s library comes with a risk score that estimates how exposed the underlying data might be to a type of attack that tries to work out whether a particular farm’s records were used in training. The authors call that score a risk-awareness indicator rather than a universal security guarantee.

The paper says the current deployment runs on Texas Advanced Computing Center infrastructure and that each farmer is assigned a server that processes only that farmer’s data and deletes it after computations finish. The code and documentation are posted publicly through the ICICLE project.

What it has not shown yet

The sandbox is a research platform, not a product farmers can download from an app store. The arXiv paper is direct about what remains to be done. It says end-to-end testing has verified the platform’s functions, from logging in and managing datasets to training and using models, but that a formal usability evaluation with farmers and agricultural stakeholders is still future work. The authors also list plans to add safeguards against misuse of the similar-farmer matching and chat features, and to study farmer onboarding and other barriers to adoption.

That is where the new USDA funding fits. The grant runs four years, and the university says the team will work directly with farmers. Whether growers trust the system enough to upload their numbers is the question the research cannot answer on paper.

Ayday sees location as an advantage. The university notes that the USDA ranked Ohio 11th among the states by agricultural value in 2025. “We are a premier research university surrounded by a massive, diverse agricultural sector,” he said. “This allows us to rapidly bridge the gap between high-level laboratory research and practical, on-the-ground deployment with regional agricultural partners.”

Why it matters beyond Ohio

The project is aimed at commercial farms, but the underlying question reaches anyone who grows food with connected tools. Smart irrigation controllers, soil sensors and plant identification apps all collect information about a yard or a field, and most users never look at where that information goes.

The Case Western team’s bet is that farmers will share more, and AI tools will get better, if the data stays in their hands. The next four years, and the farmers the team brings in, will show whether that bet pays off.