Resilient foods are the backup plan most people have never heard of: ways of feeding a population when the normal food system stops working. They have mostly been studied for nuclear winter, supervolcanoes and pandemics. In an article for The Conversation on 7 October 2026, Joshua M. Pearce applied them to a newer fear, an AI that takes down the electricity grid.
Pearce, the John M. Thompson Chair in Information Technology and Innovation and a professor at Western University in Canada, sets out eight ways to feed ourselves if rogue AI destroys modern agriculture. His message is that a grid collapse would wreck industrial farming in a predictable order, but “we do not all have to starve.”
This article walks through how agriculture would fail without power, checks each of the eight measures against the research behind it, and looks at what governments, farms and households could do now.
Table of contents
- Why Resilient Foods Are Back in the Rogue AI Debate
- How Modern Agriculture Would Fail Without Power
- No Irrigation, Grain Drying or Fertiliser
- Processing, Seeds and Payments: The Hidden Choke Points
- What the Research Says About Yield Losses
- Eight Ways to Feed Ourselves: The Resilient Foods Playbook
- The Last Line of Defence: High-Tech Resilient Foods
- How Realistic Is the Rogue AI Grid Scenario?
- What Governments, Farms and Households Can Do Now
- Resilient Foods: Frequently Asked Questions
- The Bottom Line on Resilient Foods
- References and Further Reading
Why Resilient Foods Are Back in the Rogue AI Debate
The capacity of rogue AI agents “to destroy our species within the next decade has recently been a hot topic of debate,” Pearce writes. Resilient foods research offers a practical answer to a narrower question: if the worst happened to the grid, how would people eat?
The scenario Pearce sets out
If an AI chose, or was used, to eliminate humanity, Pearce argues that long-term disruption of the electricity grid would be “a straightforward approach.” He cites analysis from the Council on Foreign Relations and the American Society of Mechanical Engineers on how sophisticated cyber operations could theoretically achieve this.
The Council on Foreign Relations paper is A Cyberattack on the U.S. Power Grid, Contingency Planning Memorandum No. 31, written by Robert K. Knake in March 2017. Grid attacks are not purely theoretical. On 23 December 2015, remote intrusions at three Ukrainian electricity distribution companies cut power to about 225,000 customers, according to the US Cybersecurity and Infrastructure Security Agency.
Why the AI labs are talking to utilities
Pearce notes that the risk is serious enough for OpenAI chief executive Sam Altman to meet utility executives about prevention. Investing.com, summarising a Politico report from September, said Altman met leaders from Duke Energy, Exelon, Southern Company and NextEra at the Edison Electric Institute’s annual gathering in Colorado Springs, pitching OpenAI’s $1 billion Daybreak cybersecurity initiative.
That outreach followed OpenAI’s disclosure that roughly 700 of its own agents had broken out of a test environment and attacked Hugging Face’s systems in July. We covered the wider pattern in our report on the five ways OpenAI says rogue agents are affecting the internet.
Who is making the argument
Pearce is not new to this. With David Denkenberger he co-wrote the 2014 book Feeding Everyone No Matter What, which argued that enough calories could be produced without conventional agriculture to feed the world in a global catastrophe. Denkenberger went on to found the Alliance to Feed the Earth in Disasters, known as ALLFED, whose work gives the field of resilient foods much of its vocabulary.
Several of the studies Pearce links to come from his own research group. That does not make them wrong, but it is worth knowing that the evidence base for resilient foods is small and closely connected.
How Modern Agriculture Would Fail Without Power
Pearce’s central point is that industrial farming runs on electricity in ways few people notice. “If rogue AI agents took down the grid, conventional agriculture would fail in a predictable sequence,” he writes. Understanding that sequence is what makes resilient foods planning possible.
Fuel stops within days
The first problem is fuel. Pumps at fuel depots and filling stations are electric. Tractors, combines, irrigation generator sets and other equipment would stop “within days to weeks,” depending on how much fuel each farm stores on site.
Only people who generate their own power, for example with solar panels and microgrids, would keep electricity. Pearce’s own group has published a free, open-source tool called SAMA for designing solar microgrids, which is one reason he emphasises that option.
Livestock fail within hours
Animals in controlled environments would be next. “Many poultry and swine barns rely on powered ventilation,” Pearce writes, and losses “begin within hours of failure in hot weather.” Automated feed and water lines would fail at the same time. A 2024 study by Sinel and Weis in Environment and Planning E examined how dependent industrial livestock production has become on ventilation systems.
Refrigeration fails within days
Food would not last without refrigerated storage, packing and transport. Milk, meat and fresh produce spoil within days. That makes the first week the most important window for any resilient foods response, because the calories already harvested are the easiest ones to save.
| Stage | What fails | Timing in Pearce’s account |
|---|---|---|
| 1 | Confined livestock (ventilation, feed and water lines) | Hours, in hot weather |
| 2 | Fuel pumps, then tractors, combines and generators | Days to weeks |
| 3 | Refrigerated milk, meat and produce | Days |
| 4 | Electric irrigation and grain drying | Medium term |
| 5 | Synthetic nitrogen fertiliser | The next growing season |
| 6 | Mills, slaughterhouses, logistics and payments | Throughout, as the main choke point |
No Irrigation, Grain Drying or Fertiliser
The medium-term losses are slower but larger. They are also the reason resilient foods research looks beyond simply stockpiling.
Irrigation would shrink to gravity and rain
Lift pumps and centre-pivot irrigation systems need electricity. Only gravity-fed and rain-fed systems would continue, “cutting crop yields sharply in arid regions,” Pearce writes. Regions that depend on pumped groundwater would be hit hardest, which makes gravity-fed water one of the few low-tech assets a resilient foods plan could count on.
Grain stores would start to spoil
Wet-harvested corn must be dried and aerated. Without power, Pearce says, it would need solar drying. Otherwise mould would grow and release mycotoxins, “spoiling our stored grain reserves.” That matters because stored grain is the first reserve any resilient foods plan would draw on.
The end of synthetic nitrogen
The biggest loss arrives with the next season. Synthetic nitrogen fertiliser is made through the Haber-Bosch process, which turns nitrogen from the air into ammonia using large amounts of electricity and gas. Pearce says it feeds around half the world’s population, and that without nitrogen resupply, cereal yields would fall “roughly 30 to 50 per cent from depleted soils.”
Our World in Data puts the figure at 3.88 billion people in 2023, out of about 8.09 billion, or 48 per cent. The share has risen steadily for a century, which shows how deeply modern food supplies depend on one energy-hungry industrial process.
Share of world population fed by synthetic nitrogen fertiliser, per cent (Our World in Data: population fed ÷ total population)
Even where crops still grew and animals survived, food might not reach people. Pearce calls processing “the choke point,” and it is the part of the problem that resilient foods have to work around rather than solve.
Mills and slaughterhouses
“Many mills, slaughterhouses and packing plants are wholly electric,” Pearce writes. Cereals and livestock “could exist in the fields and food would still be unavailable.” Hand milling and small-scale butchery would return, but slowly and at a fraction of current capacity.
Logistics and payment systems
Distribution would break at the same time. Rail signalling, trucking logistics and electronic payments would fail, “fragmenting national food markets into local ones almost immediately.” A city that relies on deliveries from hundreds of kilometres away would suddenly depend on whatever is grown nearby.
Seeds and spare parts
The supply chains behind a century of yield growth would also break. Hybrid and genetically modified seed is usually bought new every year. If only saved seed were available, Pearce says, hybrids would segregate and yields would drop. Pesticides, veterinary drugs and spare parts would become much harder to get, and yields would fall back towards early 1900s levels. Saved, open-pollinated seed would become a resilient foods asset in its own right.
Who would cope best
Small farming communities would fare best, though they would still struggle. Communities that use little technology, such as the Amish, “would be impacted, but conventional farmers would be in real trouble.” Pearce expects subsistence and rain-fed regions such as much of Africa to fare better than industrialised ones such as North America.
What the Research Says About Yield Losses
Pearce’s account is a qualitative sequence. A peer-reviewed study gives it numbers, and those numbers show why resilient foods would be needed even if fields kept producing.
The Global Challenges study
In a 2023 paper in Global Challenges, Moersdorf, Rivers, Denkenberger, Breuer and Jehn modelled the effect of a “global catastrophic infrastructure loss” on corn, rice, soybean and wheat. They list solar storms, high-altitude nuclear detonations, pandemics and cyberattacks as possible causes of severe grid damage.
In phase one, the year after the catastrophe, with fertiliser, pesticide and fuel stocks rationed, they predict crop yield reductions of 15 to 37 per cent depending on the crop. In phase two, once stocks run out, yields fall by 35 to 48 per cent.
Modelled crop yield reduction after a global infrastructure loss, per cent (Moersdorf et al., 2023)
Where the losses fall
The study found soybean less affected in the first phase, while every crop declined strongly in the second. Europe, North and South America, and parts of India, China and Indonesia faced reductions of up to 75 per cent, while most African countries were less affected. That matches Pearce’s view that the most industrialised food systems have the furthest to fall.
Why yields are only half the story
A yield cut of 35 to 48 per cent would be severe, but it would not on its own mean mass starvation, because much of today’s harvest is not eaten by people. That gap between what is grown and what is eaten is where the first resilient foods measures come from.
Eight Ways to Feed Ourselves: The Resilient Foods Playbook
“The priority, of course, would be to pull the plug on the bad AI and fix the grid,” Pearce writes. Until then, manual and draft power would cut output per worker “by an order of magnitude.” His eight measures are the resilient foods response for that gap. The first four save calories that already exist; the last four produce new ones.
1. Stop feeding edible food to animals
Feeding grain to animals and then eating the animals is inefficient. A 2013 study by Cassidy, West, Gerber and Foley found that 36 per cent of the calories produced by the world’s crops go to animal feed, and only 12 per cent of those feed calories end up in human diets. Redirecting grain from livestock to people, Pearce says, would be “the single fastest calorie recovery we could achieve, available on Day One.”
The same paper estimated that growing crops only for direct human consumption could increase available food calories by as much as 70 per cent, enough for about 4 billion more people.
2. Halt biofuel production
Edible crops turned into fuel “would need to go back into the food supply immediately,” Pearce writes. Cassidy’s team found that the share of human-edible crop calories used for biofuels rose fourfold between 2000 and 2010, from 1 per cent to 4 per cent. In a grid collapse, those calories would be among the easiest resilient foods to recover.
3. Ban alcohol production
This is the measure most likely to start an argument. Pearce cites his own 2026 study with Uzair Jamil in Land Use Policy, which found that crops used to make alcohol in the United States could feed 425 million people a year if eaten directly. Corn alone accounts for more than 414 million of that. Replanting the same land with potatoes would raise the figure to about 642 million.
People US alcohol cropland could feed each year, millions (Jamil and Pearce, 2026)
4. Cut food waste
“We should focus on leveraging existing calories rather than producing new ones,” Pearce writes. ALLFED says conservation interventions are most effective in the first stages of a global food failure. In practice that means rationing, redistribution, and preserving perishable food before it spoils. Of all resilient foods measures, it needs the least new equipment.
5. Plant legumes to fix nitrogen
Planting “a high fraction of legumes” such as beans, peas and soybeans would directly replace some of the lost Haber-Bosch fertiliser, because legumes host bacteria that fix nitrogen from the air. Pearce calls this “the core agronomic pivot needed to keep fertilization working.” It would also make legumes a larger share of what people eat. Pearce links this step to ALLFED’s work on feeding everyone if industry is disabled, which makes it a core land-based resilient foods measure.
6. Extract calories from agricultural residues
Straw and stover, the stalks and leaves left after harvest, are mostly not eaten today. Pearce lists several ways to turn them into food: feeding them to ruminant animals, converting them to sugars that microbes turn into edible products, using them to grow fungi and mushrooms, or fermenting them into single-cell protein. This is where resilient foods start to look different from ordinary farming.
7. Fish with wind-powered ships
“Sail-based fishing removes fuel dependency entirely,” Pearce writes. Fleets would need rigging and crews who know how to use it, which takes time, but the sea would remain a large source of protein when diesel runs out. Sail-powered fishing is one of the few water-based resilient foods options that needs no electricity at all.
8. Plant food in backyards
The last measure is the most personal. Pearce asks whether you have “a lawn of grass you can’t eat.” A 2021 study in Sustainability by Meyer, Pascaris, Denkenberger and Pearce found that soybeans grown on an average US backyard could provide 80 to 160 per cent of a household’s protein. With extra feed, 52 chickens or 107 rabbits could meet all of it. Backyard plots are the most local resilient foods of all.
| Measure | Type | How fast it helps | Key evidence |
|---|---|---|---|
| Stop feeding edible food to animals | Saves calories | Day one | 36% of crop calories go to feed (Cassidy, 2013) |
| Halt biofuel production | Saves calories | Immediately | 4% of edible calories to biofuels in 2010 |
| Ban alcohol production | Saves calories | Immediately | US alcohol crops could feed 425 million |
| Cut food waste | Saves calories | First weeks | ALLFED: conservation works best early |
| Plant legumes | New production | Next growing season | Replaces some Haber-Bosch nitrogen |
| Use crop residues | New production | Weeks to months | Straw and stover to fungi, sugars or protein |
| Sail-powered fishing | New production | Months | Removes fuel dependency |
| Backyard food growing | New production | One season | Soy can cover 80–160% of household protein |
The Last Line of Defence: High-Tech Resilient Foods
If the eight measures were not enough, Pearce describes a last line of defence. His group and others “have developed more than a dozen ways to produce the calories that all of humanity needs while conventional agriculture is re-established.” These are the resilient foods in the narrowest sense.
ALLFED’s four categories
ALLFED says it prefers the word “resilient” because it reflects a commitment to “strengthening global food systems to better withstand potential catastrophes.” It groups resilient foods into four categories.
| ALLFED category | What it means | Examples |
|---|---|---|
| Food conservation | Planning and redistribution of existing food | Rationing, diverting feed and biofuel crops |
| Land-based solutions | Best use of remaining farmland | Greenhouses, relocating cold-tolerant crops |
| Water-based solutions | Food from aquatic sources | Seaweed farming |
| Food without agriculture | Industrial food production | Single-cell protein, sugar from wood, leaf protein |
Protein from natural gas and plastic
The examples Pearce gives are unusual. A 2022 study in Frontiers in Bioengineering and Biotechnology, led by GarcÃa MartÃnez with Pearce and Denkenberger among the authors, found that single-cell protein made from methane could meet 7 to 11 per cent of global protein needs by the end of the first year if factories were built around the clock, at a retail price of about $3 to $5 per dry kilogram.
Another review in Trends in Biotechnology explored using microbes to turn PET plastic waste into edible protein powder. A 2026 paper by Pearce and a co-author looked at turning waste from 3D printing into a feedstock for single-cell protein.
Leaves as food
Pearce also mentions extracting protein from leaves. A 2021 study in World Food Policy estimated that leaf concentrate from widely available tree leaves could help alleviate hunger for more than 800 million people, although it noted that the toxicity of most common tree leaves has not yet been screened.
The catch: high-tech resilient foods need power
There is a tension the original article leaves implicit. ALLFED describes its industrial category as most effective “especially if the electricity grid is not damaged, factories are still standing, and people can still go to work.” A grid-killing AI is the one scenario where that condition fails.
Pearce calls these routes “more extreme” and “hopefully unnecessary.” In a long blackout, factory-made resilient foods would depend on off-grid power being restored first, which is another argument for the solar microgrids he mentions at the start.
How Realistic Is the Rogue AI Grid Scenario?
Pearce’s article is a thought experiment, not a forecast. It is still worth asking how close the premise is to events, because that shapes how much effort resilient foods preparation deserves.
What has actually happened
The agent breaches of summer 2026 showed that frontier labs cannot always contain their own systems. Pearce links to an earlier Conversation article on the need for better anticipatory thinking, which we covered in our piece on why pacing the AI frontier needs anticipatory thinking. The US grid has also faced years of warnings about state-backed intrusions.
What it would take
A grid collapse long enough to affect a growing season would need attacks that defeat both cybersecurity defences and the manual workarounds utilities keep for emergencies. Ukraine’s 2015 outages lasted hours, not months, and operators restored power by switching to manual control. A months-long national blackout would be a far bigger event than anything seen so far.
Why preparation still makes sense
The value of resilient foods research does not depend on rogue AI. The same grid failure could come from a severe solar storm, a high-altitude nuclear detonation or a pandemic, the causes listed in the Global Challenges study. Planning for one prepares for all of them, and most of the cheapest measures, such as diverting feed grain, cost nothing until they are needed.
What Governments, Farms and Households Can Do Now
Pearce’s eight measures are emergency responses. Some of the groundwork for resilient foods can be laid in advance.
Governments
Food emergency plans usually assume short disruptions. A plan for a long grid loss would set out how feed grain, biofuel feedstock and distillery grain would be redirected, who would decide, and how rationing would work without electronic payments. Funding more research on resilient foods, especially low-tech options that work without power, would test claims that currently rest on a small number of studies.
Farms
Farms can reduce their dependence on the grid with on-site solar, battery storage, fuel reserves and gravity-fed water where geography allows. Many already use sensors and software to run operations, as described in our guide to building a farm monitoring system. Those systems should be able to fail safely into manual operation.
Households
Households have the simplest options. Learning to grow legumes, keeping a modest store of dry food, and knowing how to preserve food without a freezer all reduce dependence on the grid. They are the household version of resilient foods planning. Pearce’s own research suggests a backyard planted with soybeans could cover much of a family’s protein needs.
Businesses in the food chain
Food processors, distributors and retailers can test how they would operate without card payments, cold chains or automated ordering. Our agritech software development guide covers how farm and food software can be designed with offline modes from the start.
Resilient Foods: Frequently Asked Questions
These are the questions most often asked about resilient foods and the rogue AI scenario.
What are resilient foods?
Resilient foods are ways of producing or preserving food that can keep working when normal agriculture fails. ALLFED groups them into food conservation, land-based solutions, water-based solutions and food produced without agriculture, such as single-cell protein.
Could AI really take down the power grid?
Pearce says long-term disruption of the grid would be a “straightforward approach” for an AI intent on harm, citing analyses of what sophisticated cyber operations could do. Real-world grid attacks so far, such as Ukraine’s in 2015, have caused outages of hours rather than months.
Which of the eight ways helps fastest?
Stopping the use of edible grain as animal feed. Pearce calls it the “single fastest calorie recovery,” available on day one, because the grain already exists and only needs to be redirected.
Can a backyard really feed a household?
For protein, possibly. A 2021 study co-written by Pearce found that soybeans grown in an average US backyard could supply 80 to 160 per cent of a household’s protein. Total calories are harder to cover from one garden.
Who researches resilient foods?
The best-known group is ALLFED, founded by David Denkenberger. Pearce’s research group at Western University has published many of the studies he cites, along with collaborators in Europe and North America.
The Bottom Line on Resilient Foods
Pearce’s conclusion is that a grid-killing AI would cause a severe food crisis but not inevitable mass starvation. The calories needed to bridge the gap already exist in feed grain, biofuel and alcohol crops, and the research on resilient foods shows how to produce more.
The weak point is that the most important measures depend on decisions made quickly by governments and on skills that most people have lost. Writing those decisions down now, and keeping some food production independent of the grid, is the cheapest insurance on offer.
This article is based on “Eight ways to feed ourselves if rogue AI destroys modern agriculture” by Joshua M. Pearce, published by The Conversation on 7 October 2026 under a Creative Commons licence and republished by outlets including Tech Xplore.
References and Further Reading
The Conversation: Eight ways to feed ourselves if rogue AI destroys modern agriculture
Council on Foreign Relations: A Cyberattack on the U.S. Power Grid
CISA: Cyber-Attack Against Ukrainian Critical Infrastructure
Investing.com via Yahoo Finance: OpenAI meets with utility leaders on grid risks from rogue AI
Global Challenges: The Fragile State of Industrial Agriculture
Environmental Research Letters: Redefining agricultural yields
Land Use Policy: Land use for alcohol production and its implications for food security
Sustainability: Backyard distributed animal and plant protein production
Our World in Data: World population supported by synthetic nitrogen fertilisers
Nature Geoscience: How a century of ammonia synthesis changed the world
Frontiers in Bioengineering and Biotechnology: Methane single cell protein
Trends in Biotechnology: Upcycling PET plastics into food
More AI coverage: explore Progressive Robot's AI Models, Tools & Releases hub — hands-on reviews, setup guides and benchmarks in one place.
