Farms and the food businesses behind them get hacked in four main ways. Ransomware hits the office, ERP and logistics systems at processors, grain co-ops, cold storage firms and distributors, and plants stop because nothing can be ordered, labelled or shipped. Controllers for irrigation, refrigeration and livestock get reached from the internet, often through default passwords or flaws in their web interfaces. Equipment makers' cloud portals and APIs expose machine and customer data. And, with far more effort and usually physical access, researchers have shown that the electronics inside tractors can be opened up and reprogrammed. The documented damage so far has come mostly from the first two, and most of the long-lived exposure sits in the software supply chain.
That is the short answer. This guide is the long one. It explains how precision agriculture works: satellite guidance with real-time kinematic (RTK) corrections, the ISOBUS network that lets a tractor drive an implement from another brand, telematics, farm management platforms and who owns the data they hold. It walks through the documented research on tractors, including the 2021 work by Sick Codes and colleagues on John Deere and Case IH systems and the 2022 DEF CON tractor display jailbreak. It covers right to repair and firmware locking from both sides, then farm IoT: soil sensors, irrigation controllers, milking and feeding systems, greenhouse climate computers and grain bin monitoring. It covers ransomware in food processing and grain co-ops, attacks on the cold chain, and the agtech software supply chain. It ends with detection and with hardening checklists for farmers, co-ops, processors and equipment makers.
It is written from the attacker's point of view for defenders. For each attack class you will find how it works and why, what an attacker needs, real cases, the impact, how to detect it and how to defend against it. You will not find exploit code, tool command lines or step-by-step instructions. Incidents are described by what happened and what it cost, without attributing them to any group.
A note on the law. Testing machines, controllers or cloud services that you do not own, or have no written permission to test, is illegal in most countries. Everything in this guide is for protecting your own farm, your organisation's sites and equipment with authorisation, or the products you build. The researchers cited here worked on equipment they bought, inside sanctioned events or through coordinated disclosure.
This post is part of a device security series. The companion guide on how drones get hacked and how to secure them covers mapping and spray drones in depth. The guides on how Wi-Fi gets hacked and how routers get hacked cover the farm office and yard network, and how cars and EV chargers get hacked covers CAN networks and telematics in more detail. Our posts on fabless chip designers and the software supply chain and how a laptop is built explain where the chips and SDKs inside farm controllers come from.
Contents
- How does a smart farm work, and what software does it run?
- Who owns farm data, and why does it matter for security?
- What is the attack surface of a farm and its food chain?
- What does an attacker need to hack a farm?
- What did researchers find in tractor makers' systems?
- Can GNSS and RTK guidance be jammed or spoofed?
- How exposed are ISOBUS and the machine networks?
- What does right to repair mean for tractor security?
- How does farm IoT get attacked?
- How are agricultural drones attacked?
- How has ransomware hit food processors and grain co-ops?
- How can the cold chain be attacked?
- What is in the agtech software supply chain?
- How should you weigh these risks?
- What safety systems and manual fallbacks already exist?
- How do you detect attacks on farms and food plants?
- How should farmers and growers protect themselves?
- How should co-ops and food processors secure their operations?
- How should equipment and controller makers build secure products?
- How Safeguard helps teams that build agtech and food software
- The takeaway
- Frequently asked questions
- Sources
How does a smart farm work, and what software does it run?
A smart farm runs on machines that steer themselves using satellite positioning with RTK corrections, talk to implements over the ISOBUS network, and send data through a cellular modem to the maker's cloud and a farm management platform. Around them sit sensors and controllers for water, livestock, climate and storage. Every one of these runs firmware.
Start with the tractor. A modern tractor or combine is a small network of computers called electronic control units (ECUs). Some run the engine, transmission and hydraulics on a private network that only the maker's own parts use. One acts as a gateway to the open implement network. The cab display, often called a virtual terminal, is a full embedded computer with a touchscreen, USB ports and an operating system. The tractor display that Sick Codes jailbroke in 2022 was a John Deere 4240 controller built on an NXP i.MX 6 chip running Wind River Linux 8, according to The Register, which also notes that some devices ran Windows CE.
Figure 2. The software stack inside a smart tractor and the implement behind it.
GNSS and RTK: how a tractor knows where it is to the centimetre
GNSS is the general name for satellite navigation systems such as GPS and Galileo. On its own, a GNSS receiver is accurate to a few metres. That is not good enough to plant rows or spray without overlap. Precision agriculture adds real-time kinematic positioning, or RTK. A base station at a known, fixed position measures the error in the satellite signals and sends corrections to the receiver on the machine (the "rover"). With those corrections the rover can work out its position to within a couple of centimetres. Wikipedia's RTK article lists precision farming and autopilot systems among its main uses and gives single base station accuracy figures of about 8 mm plus 1 mm per kilometre of distance from the base.
Corrections reach the machine in two common ways. One is a UHF radio link from a farm-owned or dealer-owned base station. The other is the internet, through a protocol called NTRIP (Networked Transport of RTCM via Internet Protocol). NTRIP is based on HTTP and streams correction data from a server on the internet to the receiver, usually over the machine's cellular modem. Many subscription correction services and networks of continuously operating reference stations work this way.
Autosteer then uses that position, plus an inertial measurement unit (IMU) that senses motion and heading, to keep the machine on its planned line. When the position is lost, the system does not guess. As an AEM article on interference puts it, without a reliable position autonomous machines stop and need an operator to restart them. Keep that behaviour in mind. It matters for the risk picture later.
ISOBUS: the plug-and-play network between tractor and implement
Before ISOBUS, a sprayer from one maker needed its own control box in the cab of a tractor from another. ISOBUS, formally ISO 11783, fixed that. It is a CAN-based network harmonised with the SAE J1939 standard used in trucks. Its parts define, among other things, a virtual terminal that lets any implement draw its own screens on the tractor's display, a tractor ECU that acts as a gateway between the open ISOBUS and the tractor's private network, a task controller that loads prescription maps and logs what was applied, and a file server.
CAN, the Controller Area Network, was designed for reliable communication inside vehicles. As Wikipedia's CAN bus article puts it, CAN does not support any security features intrinsically and applications are expected to add their own. In practice, any node on the bus can send any message. That design choice is common to cars, trucks and tractors, and it is why physical access to the wiring matters so much in every one of them.
Telematics, the maker cloud and farm management platforms
Most new machines have a telematics gateway with a cellular modem. It sends machine hours, location, fuel use, alarms and as-applied data to the maker's cloud. From there a farmer, a contractor or a dealer can see the fleet, move files to and from machines and, on some systems, view the cab display remotely.
Farm management information systems (FMIS) sit on top. These platforms hold field boundaries, soil tests, yield maps and prescriptions. They connect to maker clouds and to each other through APIs, and they share data with agronomists, input suppliers, co-ops, insurers and lenders. When the NEW Cooperative attack in Iowa took systems offline in 2021, its precision agriculture business SOILMAP was affected too, and The Record reported that SOILMAP's site said it integrated with major agricultural systems, John Deere among them. That is the shape of the ecosystem: many platforms, all connected.
Around the machines: farm IoT and the food chain
Off the machine, software runs the farm too. Soil moisture probes and weather stations feed irrigation controllers that start pumps and open valves. Milking robots and automated feeders record each animal's yield, health and breeding data. Greenhouses run climate computers that drive vents, screens, heating, lighting and fertigation. Grain bins carry temperature and moisture cables that drive aeration fans and dryers. Beyond the farm gate, grain elevators, co-ops, processing plants, cold storage warehouses, logistics firms and retailers run their own mix of office IT and industrial control systems.
Figure 3. Records flow up from the field, prescriptions and settings flow back down. Each hop is a trust boundary.
Who owns farm data, and why does it matter for security?
Farm data ownership is set mostly by contract, not by law. US industry principles say farmers should own the data from their operations and control how providers share it. For security, the contract decides who can see a farm's records, which partners hold standing access, and how a farmer gets data back.
The best-known framework in the US is the set of Core Principles for agricultural data. Ag Data Transparent, the body that maintains them, says the American Farm Bureau Federation worked with commodity groups, farm organisations and technology providers to set out the original Privacy and Security Principles for Farm Data in November 2014, and that members updated them in 2024. Their starting point is plain: farmers should own information that originates from their farming operations. Providers are asked to say what they collect, why, who they share it with, how it can be exported, and what happens to it when a contract ends. Ag Data Transparent runs a certification, and the principles say members "should strive" to put the principles into their contracts. They are not mandatory.
Why does this belong in a security guide? Three reasons.
First, data held is data that can leak. A yield map, a machine's location history or a herd's health record tells a buyer, competitor or thief a lot about a business. The less a provider keeps and the shorter it keeps it, the smaller the breach.
Second, sharing creates standing access. Each agronomist, co-op or app connected to a farm account through an API holds a key. Those keys rarely expire on their own.
Third, prescriptions are control data, not just records. A variable-rate prescription map tells a sprayer or spreader how much to apply on each square metre. Kaspersky's write-up of the 2021 DEF CON research raised exactly this: networked machinery could in theory be misused to damage fields, for example by altering fertiliser doses. No such attack has been documented. But a changed prescription file would do real harm, which is why the files deserve the same integrity checks as software.
Practical steps for farmers are simple. Read the data section of each platform's terms. Prefer providers that let you export everything in open formats. Review connected apps and partners each season and remove those you no longer use. And ask your provider how a prescription file is protected between the agronomist's screen and the machine.
What is the attack surface of a farm and its food chain?
The attack surface runs from field to shelf: machines and their guidance, telematics and implements; on-farm controllers for water, livestock, climate and storage; cloud platforms, APIs, dealer tools and correction services; and the office, ERP, logistics and refrigeration systems of co-ops, processors, cold stores and retailers. Most documented incidents hit the last group.
Figure 1. Every box runs software. Most incidents to date hit office and cloud systems, not the tractor.
It helps to sort the surface by how an attacker reaches it.
Remote, over the internet. Maker clouds and partner portals, farm management platforms and their APIs, NTRIP casters, and any controller whose web or programming interface is reachable from the internet. Also the office email, VPN and remote management tools at co-ops and processors, which is where most ransomware begins.
Local, within radio range. Wi-Fi and Bluetooth on cab displays and phones, wireless sensor networks in fields and barns, the UHF link from an RTK base station, and GNSS signals themselves. An attacker has to be on or near the farm.
Physical. The CAN and ISOBUS wiring, diagnostic ports, USB ports on displays, and the circuit boards inside controllers. This needs hands on the machine, which on a farm is easier at night or on a dealer lot than in a locked plant.
Supply chain. Firmware and software that arrive already flawed or tampered with, update channels, dealer service tools, and third-party software that a business has to trust, such as remote management tools or ERP systems.
What does an attacker need to hack a farm?
For most real incidents, an attacker needed only an internet connection and one weak point: a phished password, an unpatched server, a remote access tool, or a controller left online with a default password. Attacks on the machines themselves need far more: radio equipment, time near the farm, or the device on a workbench.
That split explains the incident record. Ransomware crews do not need to understand grain dryers or milking robots. They need a way into a Windows network, and the rest is routine for them. The DTN report on NEW Cooperative cites a source who told ZDNet that the attackers used employee credentials left exposed by poor password management. The Bakker Logistiek director suspected an unpatched Microsoft Exchange flaw.
Exposed controllers are next easiest. When farms in northern Israel had their irrigation controllers disrupted in April 2023, an industrial security specialist told SecurityWeek that the farms had likely left the systems exposed to the internet with default passwords, and another described the method as finding unprotected human-machine interfaces online and making changes through them. No exploit was needed.
Machine attacks sit at the other end. The 2022 tractor display jailbreak took months of work on several consoles and needed changes to the circuit board, according to The Register. Spoofing GNSS signals well enough to move one machine off its line needs radio skill, equipment and proximity. Injecting messages onto ISOBUS needs access to the wiring.
So the attacker profile is really two profiles. Financially motivated crews who hit whatever is easy and pays. And skilled researchers, mostly working in the open, who show what is possible on the machines so that makers fix it.
What did researchers find in tractor makers' systems?
In 2021 a group led by the researcher known as Sick Codes reported web and API flaws in John Deere and Case IH systems, including exposed admin tools and data about equipment and owners. In 2022 Sick Codes showed a jailbreak of a John Deere display that needed physical access. The reported issues were fixed, and Deere disputed some claims.
DEF CON 29 (2021): web portals, APIs and internal tools
Sick Codes presented the research at DEF CON 29 on 8 August 2021, and published a detailed write-up the next day with a team of about ten researchers. The research ran for about six months. The issues it reported were in the companies' web estates, not in tractors:
- Username enumeration. A Deere account API confirmed whether a username existed on each keystroke, with no rate limit. Kaspersky's summary says the team sent 1,000 queries in a couple of minutes and got 192 matches against Fortune 1000 company names.
- An equipment records tool without access checks. An internal service for recording purchased equipment did not check rights, so a user could view data on any tractor or combine, including owner and location.
- A misconfigured Pega portal. The write-up ties a Deere Pega chat service to CVE-2021-27653, a misconfiguration of the Pega Chat Access Group portal that could expose data. Kaspersky describes it as a hardcoded admin password that gave access keys to a Deere client account.
- Old open source on a supplier portal. A Deere supplier support portal ran Apache Axis, and the team cited CVE-2019-0227, a server-side request forgery flaw in Apache Axis 1.4. The CVE record notes that Axis 1.4 was last released in 2006.
- Classic web flaws. DOM-based cross-site scripting on account pages, SQL injection in an internal admin panel and in a supplier invoice form, and HTTP request smuggling on several subdomains.
- Case IH (CNH Industrial). An exposed JavaMelody monitoring server that showed session data of current users, an API that returned user details by username, and an open Sidekiq queue dashboard.
The write-up says all the issues were patched from public networks. Kaspersky notes that Deere had no dedicated channel for vulnerability reports at the time, and later launched a rewards programme. John Deere said in a statement quoted by Vice that none of the claims had enabled access to customer accounts, agronomic data, dealer accounts or sensitive personal information, and that none of the issues would have affected machines in use. Sick Codes himself did not claim the web flaws gave control of machines; he drew a conclusion about security culture and acknowledged it was an inference.
What is the lesson for defenders? The machine is only as safe as the systems around it. Dealer tools, supplier portals and support chat services hold real data about real customers, and they get less security attention than the product.
DEF CON 30 (2022): the tractor display jailbreak
A year later Sick Codes showed a jailbreak of a John Deere 4240 display and ran a farm-themed version of Doom on a tractor screen. The Register's report gives the key facts. It needed physical access and changes to the controller's electronics. Once custom software was loaded, the device ran it. All the firmware's code ran as root, the highest privilege. And Sick Codes summed up the main weakness as nothing being "encrypted or checksummed properly", adding that patching it in the field was impractical and that new devices needed to be built with proper security.
The researcher framed the work as a jailbreak, not an exploit. Its main audience was the right to repair debate, which we cover below. For security, two points matter. A device that runs whatever code it is given, with everything as root, cannot defend itself once someone has physical access. And the fix is design: secure boot that checks a signature before running code, least privilege inside the device, and signed updates.
Can GNSS and RTK guidance be jammed or spoofed?
Yes. Jamming drowns out the weak satellite signals so the receiver loses its fix, and guided machines then stop. Spoofing feeds the receiver false signals so it computes the wrong position, which is harder and needs skill and proximity. RTK correction links and NTRIP accounts are a separate target. We found no documented case of an attack steering a farm machine off its line.
How jamming works and why it works
Satellite signals arrive at the ground very weak. The AEM article notes that interference can come from cheap, low-power vehicle tracking jammers as well as far more powerful sources, and that both can cause positioning errors for autonomous farm machines. Trees, power lines and solar activity cause similar losses naturally. Ionospheric scintillation during periods of high solar activity is a familiar headache for RTK users.
What the attacker needs: a jammer and proximity. Impact: lost guidance. Machines stop or hand control back to the operator. Work is delayed, and rows planted during degraded positioning may be off. Detection: receivers report signal-to-noise drops, loss of RTK fix and jumps between fix types; logs of these events across several machines at once point to interference rather than a single faulty receiver. Defence: receivers with interference detection and multi-frequency, multi-constellation tracking; IMU integration to ride through short outages; and operating procedures that treat loss of fix as a normal event.
How spoofing works
Spoofing is the harder attack. The attacker transmits counterfeit satellite signals strong enough for the receiver to lock onto them, then shifts the computed position gradually. A careful spoofer moves the position slowly enough that inertial sensors do not flag the drift at once. Doing that to one specific moving machine, without disrupting everything around it, needs radio expertise, equipment and time near the field.
Impact: a machine that steers to the wrong line, misapplied inputs, or crop damage. Detection: disagreement between GNSS and inertial or wheel sensors, sudden changes in signal power or clock behaviour, and position jumps that do not match machine speed. Defence: receivers with spoofing detection; authenticated navigation signals where available (our explainer on whether Galileo OSNMA stops GPS spoofing covers what signal authentication does and does not do); geofences on field boundaries; and an operator present for autonomous work.
The correction link is part of the system
RTK corrections are an input that changes where the machine thinks it is. If an attacker could alter the corrections, or impersonate a base station, they could shift positions without touching the satellite signals. For UHF links, that means radio access near the farm. For NTRIP, it means the caster account and the connection.
Defensive steps are mundane but worth doing. Give each machine or user its own correction service credentials instead of one shared login. Prefer correction services and receivers that support encrypted connections, and check whether yours do. Watch for logins to your correction account from unexpected places. And keep base station firmware and any computer that serves a farm-owned correction stream up to date, because these are ordinary embedded devices and servers.
How exposed are ISOBUS and the machine networks?
ISOBUS inherits CAN's lack of authentication, so any device on the bus can send any message. That makes physical access the main barrier. Academic testbeds now study the risk, and we found no documented remote attack on ISOBUS in the field. The practical risks are a malicious device plugged into the bus and untrusted files loaded through USB.
ISO 11783 was written for interoperability and reliability. It is not a security standard, and nothing in classic CAN stops a node from claiming to be another node. This is a known property shared with automotive networks, and the car industry spent the last decade adding gateways, message authentication and intrusion detection on top.
Agricultural research on the topic is young. A research team described STAVE, a Security Testbed for Agricultural Vehicles and Environments, built to find vulnerabilities in off-the-shelf agricultural components, with initial work on wireless and CAN bus components. Their abstract makes the point that earlier work had focused on general concerns, with minimal research on finding vulnerabilities in actual agricultural systems.
What an attacker would need: access to the bus, either through a connector, a compromised implement ECU, or a compromised cab display. Impact: wrong commands to implement sections, false readings on the display, or a denial of service that stops the implement. The tractor's own engine and transmission controllers sit on its private network behind the tractor ECU gateway, which is one reason ISO 11783 defines that gateway in the first place. Detection: unexpected source addresses on the bus, message rates outside normal ranges, error frames and diagnostic trouble codes that appear only when a particular implement or device is connected. Defence: for makers, strict filtering at the tractor ECU, authenticated diagnostics and updates, and validation of every value received from an implement. For farmers, buy implements and add-on boxes from known sources, and treat aftermarket devices that plug into the bus with the same care as software you install.
USB deserves its own mention. Prescription maps, shapefiles, guidance lines and software updates often travel on USB sticks between the office PC and the cab. A display that parses files from USB is an input surface, and a stick that has been in an infected office PC is a carrier. Use dedicated sticks for machine files, scan them on a clean computer, and prefer signed update packages from the maker.
What does right to repair mean for tractor security?
Right to repair is the push for farmers and independent shops to get the tools, software and information to fix equipment. Makers and dealers argue that broad access risks tampering with safety and emissions controls and exposes intellectual property. Advocates argue that repair access and security are compatible. Both sides make points that security teams should take seriously.
Some context first. Modern machines lock some repairs behind software: a replaced part may need to be paired or calibrated with dealer tools before the machine accepts it. The Register described these as software-level locks that limit work to authorised dealers.
In January 2023 the American Farm Bureau Federation and John Deere signed a memorandum of understanding. Iowa State's Center for Agricultural Law and Taxation summarises it: Deere agreed to give farmers and independent repairers timely electronic access to manuals, software and specialty tools on "fair and reasonable terms", which may include a fee or subscription. Farmers get access to their machines' codes and data and can reset security functions disabled in the ordinary course of repair. Deere need not provide information that would override safety features or emissions controls, or disclose trade secrets. The Farm Bureau agreed to encourage its state organisations not to promote right to repair legislation. NPR reported that some farm groups doubted the deal, with the Montana Farmers Union president saying there was no commitment from anyone to enforce it.
In April 2023 Colorado became the first US state to pass an agricultural right to repair law, requiring manufacturers to provide manuals, tools, parts and software. KUNC's report captures both sides: manufacturers and dealers warned that owners could raise horsepower and bypass emissions controls, putting safety and the environment at risk, and argued trade secrets would be exposed; farmers said waiting days for a technician could cost a crop.
Where the security arguments actually sit
The maker's concern is real. A machine that will accept any firmware will also accept malicious firmware, and safety and emissions controls exist for good reasons. Security keys and dealer tools that leak become tools for thieves and tamperers.
The owner's concern is also real. Locks that depend on obscurity do not stop skilled attackers; they mostly stop the owner. A device whose security rests on nobody being able to read the firmware is weak no matter who holds the dealer tool. And, as Sick Codes' 2022 work showed, displays that check nothing can be opened by someone determined.
There is a design path that serves both. Secure boot and signed firmware protect the code that runs safety functions, so even a repairer with full diagnostic access cannot replace it with unsigned code. Diagnostic and calibration functions can be exposed through authenticated, logged interfaces available to owners and independent shops on fair terms. Parts pairing can verify genuine parts without tying repairs to one dealer. And a software bill of materials (SBOM) for each machine tells owners and repairers what is inside, including open source components and their licences. On that last point, The Register reported that iFixit's CEO and Sick Codes questioned whether Deere met its GPL source code obligations. A maker that knows exactly what open source it ships, under which licences, avoids that argument entirely.
How does farm IoT get attacked?
Farm IoT gets attacked mostly through exposure and defaults: controllers reachable from the internet, default or shared passwords, unpatched firmware, and flat networks where an infected office PC can reach the barn. Documented cases include irrigation controllers disrupted in 2023 and a Swiss dairy farm where ransomware hid herd health data.
Soil and weather sensors
Field sensors are usually low-power devices that send readings over LoRaWAN, cellular or proprietary radio to a gateway and a cloud dashboard. On their own they are low risk; a false soil moisture reading is a nuisance. The risk is in what trusts them. If irrigation runs automatically from sensor values, a tampered or failed sensor changes how much water goes on. Defence: sanity checks in the irrigation logic (does this reading match the weather and the neighbouring sensors?), alerts on sudden changes, and the vendor's account protected with multi-factor authentication (MFA).
Irrigation controllers
Irrigation controllers range from a smart timer in a shed to programmable logic controllers (PLCs) running pumps and centre pivots, often with a web or app interface for remote control. That remote access is the risk.
The 2023 case. In April 2023, automated irrigation controllers at roughly a dozen farms in the Jordan Valley and other areas of northern Israel were disrupted, forcing farmers to irrigate manually, as SecurityWeek reported. The farms had been warned to disable remote connections and had not. Industrial security specialists quoted in the article said the farms had likely left systems exposed with default passwords, that the controllers were Unitronics PLCs, and that the attackers found unprotected HMIs online and made changes through them.
The related CVE. Later that year the default password problem in Unitronics programming software was given an identifier: CVE-2023-6448 states that VisiLogic before version 9.9.00, used in Vision and Samba PLCs and HMIs, uses a default administrative password, and that an unauthenticated attacker with network access can take administrative control. Our CVE-2023-6448 explainer covers it in detail.
Impact: pumps stopped or started, schedules changed, crops stressed in hot weather, water wasted, and staff pulled into manual work. Detection: controller logins and configuration changes you did not make, schedule changes, controllers listed on internet scanning services under your farm's IP address. Defence: take the controller off the internet and reach it through a VPN or a vendor's authenticated cloud service; change every default password; update programming software and firmware; keep a written manual procedure for each pump and valve.
Livestock systems: milking robots, feeders and herd data
Milking robots, feeding systems and herd management software record each animal's milk yield, activity, health alerts and breeding status. They usually connect to a farm PC and the maker's cloud.
The Swiss case. In a case reported in 2024, a dairy farmer in Hagendorn, Switzerland, found that his computer systems, which were connected to his milking robots, had been hit by ransomware, CSO Online reported in 2024 citing the Luzerner Zeitung. His milking system maker alerted him. The attackers demanded 10,000 dollars, and he did not pay. The robots kept milking, because they can run without a computer or network connection. But the vital data was gone, including pregnancy information, and he did not spot that a calf had died in the womb. The cow could not be saved. He put the damage at around 6,000 Swiss francs, mostly vet costs and a new computer.
This case is worth reading slowly. The safety design worked: milking continued. What failed was the farm office PC, and with it the information the farmer relied on to care for his animals. Defence: keep offline copies of herd records; keep the herd management PC patched, backed up and off general web browsing and email; separate it from the office network; and know how to get critical alerts from the robot or the maker's cloud if the farm PC is down.
Greenhouse climate control
Greenhouse climate computers drive heating, vents, screens, lighting, CO2 dosing and fertigation. A failure in a cold night or a hot afternoon can destroy a crop within hours. We found no documented, public case of a greenhouse climate computer being attacked, so we will not invent one. The exposure patterns are the same as irrigation: remote access for growers and consultants, vendor maintenance connections, and flat networks shared with office PCs. Defence: put the climate computer in its own network zone, give remote users MFA and named accounts, set independent temperature alarms that do not depend on the climate computer, and know the manual override for each vent and heater.
Grain bin and storage monitoring
Grain bin monitoring uses temperature and moisture cables hanging in the grain, linked to a controller that runs aeration fans and sometimes dryers, with a cloud dashboard and phone alerts. A false "all fine" reading while grain heats, or fans switched off, risks spoiled grain. Again, no public attack case is documented, and the same controls apply: no direct internet exposure, MFA on the cloud account, alerts that reach more than one person, and occasional manual checks of the grain itself.
How are agricultural drones attacked?
Agricultural drones face the same attack classes as other drones: GNSS jamming and spoofing, attacks on the control and telemetry radio link, compromised ground station apps and laptops, and flawed firmware and cloud services. Spray drones add a safety dimension because they carry chemicals. Our drone guide covers each class in depth.
Mapping drones capture imagery for crop scouting and prescription maps. Spray drones apply products on terrain that ground machines cannot reach. Both depend on GNSS, often with RTK, and both carry firmware built from a flight controller stack, radio firmware and a ground control app. For the full treatment, see how drones get hacked and how to secure them and our teardown of how a drone is made.
Two farm-specific points. Imagery and the prescriptions built from it go into the same farm management platforms as machine data, so drone accounts deserve the same MFA and partner reviews. And spray drone operators should treat return-to-home, geofences and loss-of-link behaviour as safety settings, checked before every season.
How has ransomware hit food processors and grain co-ops?
Ransomware has stopped meat plants, grain co-ops, machinery factories, salad plants, logistics hubs and distributors. In almost every case it hit office and business systems, and companies shut down to contain it. Plants paused because orders, labels, traceability and shipping depend on those systems, not because machines were taken over.
Figure 8. The machines often still run. What stops is everything that tells them what to make and where it goes.
JBS (2021)
On 30 May 2021, ransomware hit the meat processor JBS. Transport Topics reported that all its US beef plants, which handle roughly a quarter of the American beef supply, were shut, along with slaughter operations across Australia, and one of Canada's largest beef plants was idled. JBS USA paid 11 million dollars in bitcoin. Its chief executive called it "a very difficult decision" made "to prevent any potential risk for our customers", and the company said the vast majority of facilities were already running again when it paid. Operations returned to normal within days. Our analysis of the JBS incident goes further.
NEW Cooperative and Crystal Valley (2021)
In September 2021, ransomware hit NEW Cooperative, an Iowa grain and farm supply co-op with more than 60 locations, and its precision agriculture business SOILMAP. The Record reported that the attackers demanded about 5.9 million dollars and claimed to hold business data and SOILMAP source code. In the negotiation chat, the co-op told the attackers that about 40 percent of grain production ran on its software and that 11 million animal feed schedules depended on it. Those are the co-op's own figures, made under pressure, and have not been independently verified. DTN reported that the co-op refused to pay. In the same week, Crystal Valley, a co-op in Minnesota, was also hit.
AGCO (2022)
AGCO, the maker of Fendt, Massey Ferguson, Valtra and Challenger machinery, discovered a ransomware attack on 5 May 2022. In a 16 May update, it said most affected production sites and parts operations had resumed, the rest were expected back that week, data had been taken, and it expected to make up lost production over the rest of the year. Parts operations matter to farmers: a machine waiting for a part in season is a machine not working.
Dole (2023)
In February 2023, Dole shut its systems across North America. A memo to retailers dated 10 February, reported by CNN, said: "Our plants are shut down for the day and all our shipments are on hold." Grocers in Texas and New Mexico could not stock Dole salad kits for days. Dole confirmed ransomware and said the impact on operations was limited, as The Register noted.
Logistics and distribution: Bakker Logistiek, Coop Sweden and UNFI
Food moves through logistics firms and distributors, and their outages empty shelves.
Over Easter 2021, ransomware hit Bakker Logistiek, a Dutch food logistics company. Deliveries from its warehouses in Zeewolde, Tilburg and Heerenveen stopped, the company went to pen and paper, and Albert Heijn, the largest Dutch supermarket chain, had limited prepackaged cheese, Bitdefender reported. Systems were back after six days of recovery.
In July 2021, the Swedish supermarket chain Coop closed about 500 stores when its tills stopped working. Coop was not the target: its till software supplier, Visma Esscom, used Kaseya VSA, a remote management tool, and was hit in a supply chain attack through it, BleepingComputer reported. See our Kaseya VSA supply chain analysis.
In June 2025, United Natural Foods (UNFI), a major distributor to Whole Foods, found unauthorised activity in its systems and took systems offline, pausing deliveries; grocers and bakeries reported missed deliveries and shortages, KTVU reported.
Why the plant stops when the office is hit
Food plants are tightly coupled to business systems in ways a car factory is not. Products are perishable, so there is little buffer stock. Labels and lot records are what make recalls possible and what customers expect, so a plant that cannot print labels or record lots usually holds its shipments. Slaughter and processing schedules depend on livestock arrival plans and orders. And containing ransomware usually means shutting systems down before anyone knows how far it spread.
Detection: the classic ransomware signals apply: new administrator accounts, remote management tools appearing on servers, mass file renames, backups being deleted, unusual outbound data transfers before encryption. Defence: MFA on email, VPN and remote access; patching of internet-facing servers within days; offline, tested backups; segmentation that keeps plant control networks out of reach of the office network; and manual procedures for orders, labels and shipping that have actually been practised.
How can the cold chain be attacked?
The cold chain can be attacked through the IT systems of cold storage and logistics firms, as ransomware has shown, and through the refrigeration controllers that hold temperatures. Research in 2025 and 2026 found supervisory controllers that could be taken over from the network, change set points and switch off fans while showing a normal temperature.
Business systems: Americold (2023)
Americold, which operates temperature-controlled warehouses around the world, detected a network intrusion in late April 2023. It shut its network down to contain the attack, asked customers to cancel inbound deliveries into the following week, and warned that outbound shipments would be "very limited in manual capabilities", with priority for product nearing expiry, according to a customer memo reported by BleepingComputer. The Record noted it was the company's second cyberattack after an incident in November 2020.
Refrigeration controllers: what the research showed
Commercial refrigeration in supermarkets, cold stores and distribution centres usually has two layers. Field controllers sit on each cold room or display case, read its temperature probe and switch its compressor, fans, defrost heater and lights. A supervisory controller above them, connected over an RS485 serial network usually running Modbus, gives operators a web interface for alarms, set points and temperature records, as Claroty's Team82 explains.
Figure 6. Supervisory controllers are the target. An independent temperature logger catches a display that lies.
Three pieces of research define the risk:
- Frostbyte10, Copeland E2 and E3 (2025). Armis found ten vulnerabilities in Copeland's E2 and E3 controllers, which manage refrigeration, HVAC and lighting; three were rated critical and, chained together, could give unauthenticated remote code execution as root, The Register reported. Two show how weak credential design causes trouble: CVE-2025-6519, a default admin user called "ONEDAY" whose daily password can be predicted and cannot be deleted, and CVE-2025-52549, a root Linux password generated on each boot from easy-to-find parameters. Firmware 2.31F01 fixes the E3 issues and E2 received updates, though E2 had reached end of life. Copeland told The Register that about two-thirds of North American grocery stores use its products, not all of them in scope, and there was no indication the flaws had been exploited before fixes shipped.
- Copeland XWEB Pro (2026). Team82 found 23 vulnerabilities, 21 rated high, that let an unauthenticated attacker progressively bypass protections and reach root code execution, starting with an authentication bypass, CVE-2026-25085. In a physical test setup, they showed that a compromised supervisory controller could set the displayed temperature on a field controller to any value and switch the compressor, fans and defrost on or off; in one test the display kept showing the real probe reading while the fans were off, so the unit warmed with no visible sign. Copeland fixed the issues in firmware 1.13.
- Danfoss AK-SM 800A (2026). Team82 found a hidden "code-of-the-day" login mechanism that could bypass authentication (CVE-2025-41450), a command injection leading to remote code execution, and a web server configuration injection. Using internet scanning services, they counted 2,765 exposed devices. Danfoss addressed the issues in firmware, and Claroty recommends R4.3.1 or later.
These are not tractor problems, but they are food problems. A distribution centre or processor's cold rooms protect stock worth far more than the controller, and spoiled food can become a food safety issue if nobody notices. See our coverage of European supermarket refrigeration attacks and the Danfoss and Copeland research for more.
What an attacker needs: a reachable management interface. Impact: spoiled stock, product recalls if temperature excursions go unnoticed, regulatory record gaps, and in the worst case risk to consumers. Detection: set point, schedule and user changes on the controller; logins from new addresses; and, most important, an independent temperature logger with its own alarm path that will disagree with a controller that is lying. Defence: never expose management interfaces to the internet; reach them through a VPN or remote access gateway; apply the fixed firmware; change default accounts and disable maintenance back doors where the vendor allows; and put refrigeration systems in their own network zone.
What is in the agtech software supply chain?
The agtech software supply chain runs from chip vendor SDKs and embedded operating systems, through open source libraries and ISOBUS stacks, to supplier ECUs, the maker's firmware, apps and cloud, dealer tools and update channels. For co-ops and processors it also includes business software suppliers. Documented incidents have exploited several of these links.
Figure 9. Every layer is someone else's code. Each example on the right is cited in the article.
Look at the examples the public record already gives us.
Embedded operating systems. The display jailbroken at DEF CON 30 ran Wind River Linux 8. Every embedded Linux image carries a kernel, a C library, a web server or two, TLS libraries and dozens of other packages. Each has its own vulnerabilities and update cadence.
Old open source. A Deere supplier portal in 2021 ran Apache Axis 1.4, last released in 2006, with CVE-2019-0227. Old components linger in web estates and in firmware because nobody has an inventory that says they are there.
Licences. The GPL compliance questions raised in 2022 are a supply chain problem too. Licence obligations travel with the code.
Remote management. Coop Sweden closed hundreds of stores because a supplier's supplier was hit through Kaseya VSA. A business inherits the risk of the tools its vendors use to manage its systems.
Software updates. In June 2017, destructive malware spread through a compromised update of M.E.Doc, a Ukrainian accounting package, as Wikipedia's account summarises. It spread far beyond its first victims. Mondelez, the Oreo maker, reported a 5 percent drop in quarterly sales and blamed shipping and invoicing delays from the attack, Fortune reported. A trusted update channel became the delivery route.
Platform code. After the NEW Cooperative attack, the attackers claimed to hold SOILMAP source code. Stolen source code helps an attacker find flaws in a platform that many farms connect to.
Controllers in the food chain. The Copeland and Danfoss research found hidden accounts, predictable passwords and injection flaws in firmware that runs cold rooms worldwide. Fixes exist, but they have to travel through installers and service companies to reach each site.
Why is this hard for agriculture specifically? Machines stay in service for decades. Implements from many brands share one bus. Dealers sit between makers and farmers, and updates often depend on a dealer visit or a USB stick. Controllers for irrigation, livestock and refrigeration are installed by local contractors who may not be on the maker's update notices. All of which means the maker's knowledge of what is inside each product, and its ability to ship fixes quickly, decides how long a flaw stays live in the field. Our posts on SBOMs for agricultural machinery firmware, the agriculture and food tech supply chain and the firmware supply chain security guide go deeper.
How should you weigh these risks?
Start with what has actually happened: ransomware on business systems and internet-exposed controllers with weak or default credentials. Then cloud and API flaws, supplier compromise and stolen accounts. Treat GNSS spoofing and in-vehicle network attacks as real but rarer, needing skill, proximity or physical access, and design them out over equipment lifecycles.
Figure 5. Placement is a qualitative judgement from the documented cases. Start in the top left.
Figure 4. Documented incidents and research. Most stopped operations through business systems.
Three observations from the record.
The incidents that stopped food production all started in business IT. JBS, Dole, AGCO, the grain co-ops, Americold, Bakker Logistiek, Coop Sweden and UNFI were all IT or supply chain incidents with physical consequences.
The incidents that touched farms directly were small and preventable. The irrigation disruption came from exposed controllers with default passwords. The Swiss dairy case came from a compromised farm PC.
The research on machines and controllers shows what could come next. The tractor research, the ISOBUS testbeds and the refrigeration controller work all point to the same weaknesses: no secure boot, everything as root, hidden or predictable credentials, and management interfaces on the internet. Fixing those is slow because equipment lasts. That is why it needs to start now.
What safety systems and manual fallbacks already exist?
Farms and food plants have more fallbacks than headlines suggest. Guided machines stop when they lose position, operators can steer by hand, milking robots can run without a network, irrigation and refrigeration can be run manually, and plants have paper procedures. The weakness is that fallbacks are rarely practised and depend on records that may be locked.
This matters for a calm assessment. A tractor that loses GNSS does not drive into a ditch; it stops and asks for its operator. The tractor's engine and transmission controllers sit on a private network behind a gateway. The Swiss milking robots kept milking with the farm PC down. Irrigation pumps and valves have manual controls. Refrigeration field controllers keep running their last settings if the supervisory controller fails. Bakker Logistiek went to pen and paper; Americold kept limited manual outbound loads.
Where fallbacks break down is information. The dairy farmer lost his herd's pregnancy data. Plants cannot ship without traceability records. A greenhouse manager cannot set a manual vent position correctly without knowing what the climate computer was doing. So the most useful preparation is unglamorous: offline copies of the records you need to run by hand, written manual procedures at each critical machine, and a drill once a season.
How do you detect attacks on farms and food plants?
Detection starts with an inventory, because you cannot notice an unexpected change on a device you do not know exists. Then watch logins and remote access, configuration and set point changes, traffic crossing between network zones, and signs of ransomware staging. On machines, watch positioning quality and unexpected bus devices.
Signals for farmers and growers
- Login alerts from your farm platform, maker cloud, correction service or controller app that you did not trigger. Turn these alerts on.
- Schedule, set point or prescription changes you did not make.
- A controller, camera or router on your farm's IP address showing up on an internet scanning service. Your internet provider or a local IT firm can check this for you.
- Repeated loss of RTK fix across several machines at once, especially in one place.
- The farm PC slowing down, files renamed with odd extensions, or a ransom note. Disconnect it from the network at once and call for help.
Signals for co-ops, processors and cold stores
- New administrator accounts, remote management tools appearing on servers, backups or shadow copies being deleted, and large outbound transfers. These usually come before encryption.
- Traffic between the business network and Level 1 or 2 control devices that is not on your allowed list (see the next section).
- Remote access sessions outside approved windows or from new places.
- Set point, recipe or schedule changes on refrigeration, process and packaging controllers, compared against a known-good baseline.
- Disagreement between independent temperature loggers and controller displays.
Signals for equipment and controller makers
- Spikes in failed logins and enumeration patterns on customer portals and APIs.
- API calls where the user requesting a machine's data is not linked to that machine.
- Firmware images in the field whose hashes do not match what you signed.
- Vulnerability reports arriving through social media because you have no disclosure channel. Fix that first.
How should farmers and growers protect themselves?
Update displays, receivers, controllers and apps; turn on MFA for every farm platform; review connected apps each season; take controllers off the internet and change every default password; separate the office from the barn; keep offline copies of records; and practise running things by hand.
Figure 11. Short checklists for the three groups. The sections below explain each item.
The checklist in more detail:
- Update. Cab displays, GNSS receivers, telematics, drones, irrigation and climate controllers, and the apps that run them all get firmware and software updates. Ask your dealer and installers to apply them, and ask how you will hear about security fixes.
- MFA on every account. Maker clouds, farm management platforms, correction services, drone accounts, herd software and banking. One stolen password can reach a whole fleet's data.
- Review who has access. Each season, list the agronomists, contractors, co-ops and apps connected to your accounts. Remove those you no longer use. Use named accounts for staff, not one shared login.
- Nothing on the internet. No controller, PLC or HMI should be reachable directly from the internet. If you need remote access, use the vendor's authenticated cloud service or a VPN, with MFA.
- Change every default password. On controllers, routers, cameras and base stations. Write the new ones down somewhere safe and offline.
- Separate the office from the barn. Put controllers, milking systems and climate computers on a separate network from the office PC, guest Wi-Fi and family devices. A basic business router can do this; our router guide shows how.
- Offline copies of records. Herd health and breeding records, field boundaries and prescriptions, and controller settings. Keep a copy on a drive that is not plugged in.
- Practise manual operation. Know how to run each pump, vent, fan and milking unit by hand, and keep the steps written next to the machine.
- Treat USB sticks as tools. Keep dedicated sticks for machine files and do not use them in other computers.
- Report problems. If something looks wrong, tell your dealer or installer and the maker. They need to know.
How should co-ops and food processors secure their operations?
Build an asset inventory across IT and operational technology; segment the network using the Purdue model and IEC 62443 zones and conduits; secure remote access; monitor the conduits; keep offline, tested backups; and maintain an incident response plan with manual procedures and alternative supply.
Asset inventory
List every server, PC, controller, PLC, HMI, network device and cloud service, with firmware and software versions, owner, location and network zone. Include contractor-installed systems such as refrigeration, packaging lines and weighbridges. Our guide to building an OT asset inventory walks through it. Without the inventory, you cannot patch, monitor or segment with confidence.
Segmentation with the Purdue model and IEC 62443
The Purdue model, from the Purdue Enterprise Reference Architecture of the 1990s, describes industrial networks in levels. Level 0 is the physical process: sensors, valves, pumps, compressors, conveyors. Level 1 is basic control: PLCs and field controllers. Level 2 is supervisory control: HMIs, SCADA and supervisory controllers. Level 3 is site operations: historians, batch and recipe servers. Levels 4 and 5 are the business network and the cloud. An industrial demilitarised zone (DMZ) sits between Level 3 and Level 4 as the only meeting point.
ISA/IEC 62443, the main series of standards for industrial automation and control system security, adds zones and conduits. A zone groups assets with similar risk and security needs, such as all refrigeration controllers on a site. A conduit is the only allowed path between two zones, with a firewall and a written list of the flows it permits. The standards also cover security programmes for asset owners, patch management, risk assessment for system design, and secure development requirements for product makers.
Figure 7. Purdue levels give the layers. IEC 62443 zones and conduits decide what may talk to what.
For a food plant, sensible zones include the packaging lines, the refrigeration plant, wastewater and utilities, and each slaughter or processing line. For a grain co-op, the elevators, dryers and scales. Each gets its own conduit to Level 3 and nothing straight to the business network. Our posts on how to segment an OT network, the Purdue model and where it falls short and vulnerability management under IEC 62443 go further.
Secure remote access
Vendors, integrators and refrigeration service companies need remote access. Give it through one gateway in the DMZ, with MFA, named accounts, approval for each session, time limits and recording. Remove standing VPN accounts. Never leave a controller's own web interface exposed for convenience.
Monitoring
Watch the conduits. Passive network monitoring tools designed for industrial protocols can learn normal flows and flag new ones. Send controller logs, remote access logs and Windows security logs to one place. Alert on the ransomware staging signals listed earlier.
Backups and recovery
Keep offline or immutable backups of servers, PLC logic, controller configurations and recipes, and test restores. Decide the restore order in advance: identity first, then ERP, labelling and traceability, shipping, then the rest.
Incident response
Write the plan before you need it. Include who decides to shut systems down, how to contact customers and suppliers without email, manual procedures for orders, labels, lot records and shipping, and pre-agreed alternative suppliers, sites and routes for critical products. Rehearse it once a year with a tabletop exercise.
Figure 10. Each layer catches what the one above it missed. Recovery is the last line.
How should equipment and controller makers build secure products?
Ship an SBOM for every firmware and app; use secure boot and signed updates; remove default, shared, hidden and predictable credentials; run least privilege inside devices; check authorisation on every API call; publish a vulnerability disclosure route; scan code and shipped binaries; patch open source quickly; and design repair access with security, not against it.
Devices and machines
- Secure boot and signed firmware. The device verifies a signature before it runs any code. This is the direct answer to the 2022 jailbreak, where the researcher said nothing was checksummed or encrypted properly.
- Least privilege. Not everything runs as root. Separate the user interface, network services and control functions.
- No default or hidden credentials. No "code of the day", no predictable daily or boot-time passwords, no shared admin password across a product line. Force a unique credential at installation. The Copeland, Danfoss and Unitronics issues all sit here.
- Filter at the gateway. The tractor ECU should pass only what an implement legitimately needs to send, and validate every value.
- Authenticated diagnostics and updates. Diagnostic sessions and software updates require authentication and are logged, whether the user is a dealer or an independent shop.
- Safe failure. Loss of position, loss of link and unexpected input all lead to a safe stop, never a guess.
Cloud, apps and portals
- Authorisation on every request. The 2021 equipment records tool showed what happens when an API checks who you are but not what you may see.
- Rate limits and enumeration protection on login and lookup endpoints.
- Retire old components. A 2006 release of Apache Axis on a supplier portal is an inventory failure before it is a patching failure.
- Treat internal tools as external. Admin panels, monitoring dashboards and job queues should never be reachable from the internet.
Supply chain, disclosure and repair
- An SBOM for every product version, covering firmware, apps and cloud services, including licences.
- A public vulnerability disclosure policy with a working contact, safe harbour for good-faith research, and published advisories. The 2021 researchers struggled to find who to tell.
- Patch open source quickly, and have a way to ship fixed components when upstream is slow.
- Updates that reach the field. Over-the-air where possible, with clear notices to dealers and installers when not.
- Repair access designed with security. Authenticated, logged diagnostic access available on fair terms protects safety functions better than secrecy does.
How Safeguard helps teams that build agtech and food software
Safeguard covers the software supply chain inside agtech and food products in one platform: SBOMs for firmware, apps and cloud services, binary and image scanning, vulnerabilities ranked by reachability, SCA, SAST and secrets scanning on code you own, fix pull requests gated by your CI, Gold patched packages, supplier risk and compliance evidence.
Safeguard does not sit on the ISOBUS wiring or watch a cold store's network. It sits where most long-lived exposure in this guide comes from: the firmware, operating system images, open source components, apps and cloud services that equipment makers, controller makers and agtech platforms build and ship, and that co-ops and processors buy. Teams that take this seriously often end up with one tool for source scanning, another for binary analysis, another for SBOM management, a questionnaire tool for suppliers and a GRC platform for audits. Safeguard does that work end to end, with one inventory and one vulnerability model, from the first SBOM to the merged fix and the audit record.
Figure 12. How Safeguard fits an agtech, controller or food software product, from inputs to fixes and evidence.
See inside the binaries and images you ship and receive
In the Safeguard console, Integrations > Upload App Binary accepts compiled software: Windows EXE, MSI and DLL files, Android APKs and iOS IPAs, VM and disk images in QCOW2, VMDK, VHD and OVA formats, and container tarballs. Safeguard unpacks the file, identifies embedded libraries and their versions, maps them to known vulnerabilities and records licences, and the result becomes a project with a full SBOM. For agtech and food teams that covers farmer and dealer apps on both mobile platforms, Windows service and diagnostic tools, Linux images for displays, gateways and supervisory controllers, and the containers behind farm platforms. Raw firmware images for microcontrollers may need to be extracted into their component files first, and the usual limits of binary analysis apply: heavy obfuscation, encryption or unusual packing can give incomplete results. Co-ops and processors can use the same feature to look inside software they buy. See our guide to scanning app binaries with Safeguard and our explainer on binary composition analysis.
Rank vulnerabilities by what can actually be reached
A display image or controller firmware can contain hundreds of components, and many vulnerabilities in them sit in code the product never calls. Safeguard's reachability analysis ranks the vulnerabilities your code can actually reach above those it cannot, and adds EPSS (the Exploit Prediction Scoring System) and known exploited context. That is how a team tells the web interface authentication flaw that needs fixing this week apart from an unused function in a bundled library.
Scan the code you own
For display software, controller firmware, telematics services, farm platform APIs and mobile apps that you write, Safeguard runs software composition analysis (SCA) across the full transitive dependency tree, static application security testing (SAST) on your own code, and secrets scanning that catches signing keys, API tokens and hard-coded credentials before they ship. Several issues in this guide are classes these scans exist to catch: missing authorisation checks, SQL injection and command injection, hard-coded or predictable credentials, and outdated components such as a 2006 Apache Axis release.
SBOMs your customers, dealers and auditors can use
Every project produces an SBOM you can export in CycloneDX or SPDX, among other formats. That is what a co-op, processor or large farm can ask an equipment or controller maker for, and what lets a maker answer "which products ship this library" in minutes when an advisory lands. It also records licences, which settles open source compliance questions with evidence. Safeguard's Third-Party Risk Manager lets you request SBOMs from your own suppliers, score them and monitor them continuously. See SBOMs for firmware and IoT devices.
Fix it without waiting for upstream
This is where Safeguard differs most from a scanner. For code in your repositories, Griffin, Safeguard's remediation model, opens a fix pull request that your own CI builds and tests before anyone merges it, and your policies decide what can merge. For open source components where upstream has not shipped a fix, or where the fix sits in a version you cannot move to without a larger change, Safeguard Gold provides patched, hardened builds of those packages, and Safeguard can fork and patch the vulnerable package so your next build pulls a fixed version. For a controller maker waiting on an upstream release, or an equipment maker waiting on a supplier, that is the difference between shipping the fix in the next update and waiting a season. Closed source components still have to come from their vendor, but the open source around them no longer has to wait.
Compliance evidence from the same data
Safeguard's compliance suite collects evidence for frameworks such as SOC 2 and ISO 27001 from the same scans, policies and fix history. For makers working toward IEC 62443-4-1 secure development requirements, or preparing for the EU Cyber Resilience Act, the SBOMs, vulnerability monitoring records, risk decisions and fix trail are the raw material that customer reviews and auditors ask for.
Deploy where your code lives
Firmware source, signing infrastructure and customer data are sensitive. Safeguard runs as SaaS, in your own VPC, or fully on premises. It can run in Zero AI mode with no language model in the loop, or with a model you choose.
How this compares
Food and agriculture teams often combine several tools: an industrial network monitoring platform such as Claroty, Armis or Nozomi Networks to watch plant traffic and find exposed devices, a source SCA tool such as Snyk or Black Duck, a firmware analysis product such as Finite State, and a GRC tool for audits. Those are capable products, and network monitoring does a job Safeguard does not: watching live plant and controller traffic. Claroty's and Armis's refrigeration research cited above is a good example of what that side of the industry contributes. Safeguard's case is consolidation and remediation across the software itself: one inventory across source, binaries and images; one vulnerability model with reachability; fix pull requests and patched packages instead of tickets; and supplier risk and compliance evidence from the same data. For more context, see agricultural machinery software supply chain and physical safety and food processing lines software supply chain and physical safety.
The takeaway
Tractor hacking makes good headlines, and the DEF CON research earned its attention. But the attacks that have cost farms and food companies real money are plainer. A co-op's office network locked during the grain season. A meat processor's plants paused while IT systems were rebuilt. Irrigation controllers left online with default passwords. A dairy farmer's PC encrypted, taking the herd's health records with it. A cold store running outbound loads by hand.
The fixes are plain too. Know what you run. Keep controllers off the internet. Turn on MFA. Separate the office from the barn and the plant. Keep offline records and practise manual operation. And for makers, ship products whose software you can account for line by line, sign what runs, remove the hidden passwords, and be able to fix what you find without waiting for everyone upstream. The machines are built to stop safely. The job now is making sure the software around them is built to the same standard.
Try it yourself
Click through these interactive Safeguard demos to see the steps described in this guide.
Test yourself
Why did ransomware stop meat and food plants such as JBS and Dole?
The ransomware hit business IT systems, not the processing machines. Companies shut systems down to contain it, and plants paused because orders, labels, lot traceability, scheduling and shipping depend on those systems.
What did researchers find at DEF CON 29 and DEF CON 30 about John Deere?
In 2021 Sick Codes and colleagues reported web and API flaws in Deere and Case IH systems, including exposed admin tools and an equipment records tool without access checks. In 2022 a display jailbreak was shown, but it needed physical access.
What is the difference between jamming and spoofing GNSS guidance?
Jamming drowns out the weak satellite signals so the receiver loses its fix and guided machines stop. Spoofing feeds false signals so the receiver computes the wrong position, which is harder and needs skill and proximity.
What went wrong with the irrigation controllers disrupted in Israel in April 2023?
The farms had been warned to disable remote connections and had not. Specialists said the Unitronics controllers were likely left exposed online with default passwords, so attackers found unprotected HMIs and made changes through them.
Why is physical access the main barrier on ISOBUS?
ISOBUS inherits CAN's lack of authentication, so any device on the bus can send any message. The practical risks are a malicious device plugged into the bus and untrusted files loaded through USB.
How can a cyberattack spoil food in cold storage, and what is a simple defence?
Researchers showed some supervisory refrigeration controllers could be taken over from the network, letting an attacker change set points or switch off fans while the display showed a normal temperature. Keep controllers off the internet, apply updates, and use an independent temperature logger with its own alarms.
Frequently asked questions
Can a tractor be hacked?
Yes, but the documented cases needed physical access or targeted the maker's web systems rather than the machine itself. In 2022 a researcher jailbroke a John Deere display by modifying its circuit board. In 2021 researchers found flaws in Deere and Case IH web portals and APIs that exposed equipment and user data. We found no public record of a remote takeover of a tractor in the field.
What happened with the John Deere hacks at DEF CON?
At DEF CON 29 in 2021, Sick Codes and colleagues reported web and API flaws in John Deere and Case IH systems, including username enumeration, an equipment records tool without access checks, and exposed admin tools. They were patched, and Deere said none gave access to customer accounts or machines. At DEF CON 30 in 2022, Sick Codes showed a display jailbreak that needed physical access.
Can GPS guidance on a tractor be jammed or spoofed?
Jamming is easy with cheap jammers and causes guided machines to lose position and stop. Spoofing, which feeds false signals so the receiver computes the wrong position, is much harder and needs skill and proximity. Defences include receivers with interference and spoofing detection, inertial sensors, geofences, protected RTK correction accounts and an operator present for autonomous work.
Does right to repair make tractors less secure?
Not necessarily. Makers worry that broad access lets owners tamper with safety and emissions controls and exposes intellectual property. Advocates argue that security through secrecy mostly blocks owners, not skilled attackers. Secure boot, signed firmware and authenticated, logged diagnostic access can protect safety functions while giving owners and independent shops the tools they need.
How do I secure irrigation and greenhouse controllers?
Take them off the internet and reach them through a VPN or the vendor's authenticated cloud service with MFA. Change every default password, update firmware and programming software, and put controllers on a separate network from office PCs. Keep independent alarms, written manual procedures, and offline copies of settings so you can run by hand if a controller fails.
Why did ransomware stop meat and food plants?
Ransomware at JBS, Dole and others hit business IT systems, not the processing machines. Companies shut systems down to contain it, and plants paused because orders, labels, lot traceability, scheduling and shipping depend on those systems. A plant that cannot label products or trace lots usually holds its shipments. Offline backups, segmentation and practised manual procedures shorten outages.
How can a cyberattack spoil food in cold storage?
Researchers have shown that some supervisory refrigeration controllers could be taken over from the network through authentication flaws or predictable passwords. An attacker could change set points or switch off fans while the display showed a normal temperature. Fixed firmware is available. Keep controllers off the internet, apply updates, and use an independent temperature logger with its own alarms.
Who owns the data my farm equipment collects?
It depends mostly on your contracts. US industry principles maintained by Ag Data Transparent say farmers should own data from their operations and control how providers use and share it, but they are guidelines, not law. Read each platform's data terms, prefer providers that let you export everything, and review connected apps and partners every season.
Sources
Every source below was fetched and checked while writing this guide. Incidents are described by their technical and business impact, without attribution to any group.
- Sick Codes et al., Being root on two agriculture companies in good faith (August 2021)
- Kaspersky, Farm equipment security at DEF CON 29 (October 2021)
- The Register, Doom on a John Deere tractor at DEF CON 30 (August 2022)
- The Record, Q and A: Sick Codes talks tractor hacks (September 2022)
- Vice, Hackers uncover weaknesses in agriculture giants' systems (August 2021)
- CVE Program, CVE-2021-27653 and CVE-2019-0227
- Wikipedia, CAN bus, ISO 11783, Real-time kinematic positioning, NTRIP and Purdue Enterprise Reference Architecture
- AEM, How natural phenomena and intentional jamming can interfere with the autonomy of agricultural machinery (June 2024)
- Freyhof, Grispos, Pitla and Stolle, Towards a Cybersecurity Testbed for Agricultural Vehicles and Environments (arXiv, 2022)
- Ag Data Transparent, Core Principles
- Iowa State CALT, American Farm Bureau and John Deere sign MOU on right to repair (January 2023)
- NPR, Right to repair advocates skeptical of John Deere agreement (January 2023)
- KUNC, Colorado becomes first to pass right to repair for farmers (April 2023)
- SecurityWeek, Irrigation systems disrupted by attacks on industrial controllers (April 2023)
- CVE Program, CVE-2023-6448
- CSO Online, Ransomware attack paralyzes milking robots (August 2024)
- Transport Topics, JBS paid hackers 11 million dollars after hack closed plants (June 2021)
- The Record, Ransomware attacks on grain co-ops (September 2021)
- DTN, Cooperative won't pay hacker ransom (October 2021)
- AGCO, Update on recovery from ransomware cyber attack (May 2022)
- ABC17 News (CNN), Cyberattack on Dole temporarily shuts down North America production (February 2023)
- The Register, Dole production plants crippled by ransomware (February 2023)
- Bitdefender, Ransomware attack causes supermarket cheese shortage in the Netherlands (April 2021)
- Threatpost, Ransomware attack creates cheese shortages in Netherlands (April 2021)
- BleepingComputer, Coop supermarket closes 500 stores after Kaseya ransomware attack (July 2021)
- KTVU, Whole Foods working to restock shelves as distributor faces cyberattack (June 2025)
- Fortune, Cyber attacks hit earnings at FedEx, Mondelez and Merck (August 2017)
- Wikipedia, 2017 Ukraine ransomware attacks (for the M.E.Doc update channel)
- BleepingComputer, Cold storage giant Americold outage caused by network breach (April 2023)
- The Record, Cold storage company Americold reports cyberattack (May 2023)
- The Register, Frostbyte10 bugs put thousands of refrigerators at major grocery chains at risk (September 2025)
- CVE Program, CVE-2025-6519, CVE-2025-52549, CVE-2026-25085 and CVE-2025-41450
- Claroty Team82, Chilling discoveries: vulnerabilities in Copeland XWEB Pro controllers (August 2026)
- Claroty Team82, Freeze the controller, defrost the food: Danfoss AK-SM 800A (August 2026)
- ISA, ISA/IEC 62443 series of standards