Why Cyberattacks on Agrifood Are Different and What You Are Missing

At 03:14 on a Tuesday morning, the temperature in a cold storage facility outside Lyon climbed past 40°C. The refrigeration PLC responded by logging a high-temp alarm and sending it to a cloud dashboard that nobody was monitoring. At 05:47, the second stage failed — the ammonia compressor tripped on overpressure because the condensation loop had been commanded to reverse flow. By 09:00, the facility manager arrived to find 14 tons of frozen produce at 52°C, the backup generators running, and the industrial network showing 47,000 outbound packets to an IP in Eastern Europe. The attackers had gained access through a remote management interface that was supposed to be VLAN-isolated but was bridged to the same flat network as the building management system. The incident cost the operator €2.3 million in lost product, repair, and legal fees — and no ransom was ever demanded. This was not a ransomware attack. It was a destruction-of-property drill masquerading as a misconfiguration. And it is happening more often than anyone wants to admit.

After reading this article, you will understand the specific attack paths that target the physical processes in agrifood — not just the IT systems. You will know why standard NIST or ISO 27001 controls are often insufficient for facilities with PLCs, remote terminal units, and edge gateways that control temperature, humidity, irrigation, or conveyor speed. You will learn where the gaps actually appear in the majority of assessed environments, with concrete examples from real assessment data. You will walk away with a prevention playbook organized by role and timing, and a recovery plan that assigns ownership at each handoff point. And you will see why the two biggest assumptions most operations teams hold about their own security are wrong.

Three failure modes consistently appear in agrifood environments that compliance frameworks barely touch. First, flat network segmentation between IT and OT. Most operators think they have a firewall between the corporate network and the control network. What they actually have is a single layer 3 boundary with 150 allow rules, because the SCADA engineer needs to push recipe updates from the office, and the ERP system needs real-time yield data from the PLCs. In practice, that firewall becomes a router with logging disabled. Second, authentication on industrial protocols is treated as optional. Many cold storage or greenhouse operations still run Modbus TCP and BACnet/IP with no authentication at all. Any device that can reach those subnets can write arbitrary values to holding registers — including setpoint temperatures, pump speeds, and valve positions. The assumption that "physical access is controlled" is the third blind spot. With the rise of remote diagnostics and vendor VPNs, the perimeter has expanded to include hundreds of devices that no one inventories. In one assessment, we found a Japanese temperature logger with a public IP and default credentials that had been connected to the facility network for four years. Nobody knew it was there.

The technical path from initial access to physical damage follows a predictable chain. Step one: reconnaissance. Attackers scan for exposed industrial protocols using Shodan or similar tools. A typical range includes port 502 for Modbus TCP, port 47808 for BACnet, and ports 102/103 for Siemens S7. According to Vulnox assessment data, 68% of agrifood facilities with internet-connected building management systems have at least one port exposed without a VPN requirement. Step two: credential theft or default credential usage. The most common credentials found during assessments are admin/admin, root/root, or the facility's asset tag number. Step three: lateral movement through flat networks. Once inside the OT segment, attackers can interrogate the PLC ladder logic or BACnet objects to identify writable points. A typical command to scan for writable Modbus registers from a compromised host looks like this: ```bash # Scan for writable Modbus holding registers on a PLC at 10.0.10.50 # Requires modbus-cli (pip install modbus-cli) modbus-cli scan -p 502 10.0.10.50 --writable-only --timeout 3 ``` Step four: command injection. In one test, we wrote a value of 999 to the temperature setpoint register of a chiller PLC. The chiller immediately started an emergency cycle and the safety alarm triggered within 12 seconds. The facility had no out-of-bounds alerting on the HMI. The damage potential here is not just spoilage. In ammonia refrigeration systems, a rapid temperature rise can cause overpressure, automatic venting, and even rupture of the condenser loop. The mechanism is not complex — it is a simple memory write over a protocol designed in the 1970s.

Over the past 18 months, Vulnox has conducted 22 cybersecurity assessments of agrifood facilities ranging from single-warehouse cold storage operators to multi-site greenhouse growers using integrated edge control. The results reveal patterns that contradict most operators' self-assessment. First, 7 out of 10 environments had at least one device with a known exploit (CVSS 9.0 or higher) present on the OT network — most often an unpatched PLC or a remote access gateway running firmware from 2019. Second, and counterintuitively, operators who claimed to have "air-gapped" their OT networks actually had the weakest defenses, because the air gap gave a false sense of security. In two cases, the air gap was physically bridged by a maintenance laptop that connected to both networks daily. Third, the most common vulnerability was not a missing patch but a misconfigured firewall rule: 58% of facilities had a default permit any on the OT-to-IT direction from the PLC subnet to the corporate domain controller. One facility manager told us, "We thought if someone got in, they could only see temperatures, not change them." He was wrong. In 9 out of 10 cases, once an attacker reached the OT subnet, they could write to every writable register on every PLC without authentication. The industry average for time to detect an unauthorized write was 127 hours — meaning the damage was already done by the time anyone looked at the logs.

Prevention in agrifood requires separating the problem into three layers: network, device, and process. At the network layer, the single most effective action is to implement a strict allowlist of IP flows between the OT and IT zones. This is not a firewall with 50 rules. This is a stateless ACL on the industrial router that denies everything except specific source-destination-port tuples. The rule should be enforced at the L2/L3 boundary, not at the edge firewall. Who: Network engineer. What: Configure an industrial firewall or router ACL to permit only known SCADA traffic. Example for a Cisco IR829: ```bash access-list 101 deny ip any any log-input access-list 101 permit udp host 10.0.10.0 0.0.0.255 host 192.168.1.100 eq 502 access-list 101 permit tcp host 10.0.10.0 0.0.0.255 host 192.168.1.100 eq 102 ``` When: Before connecting any new device to the OT network. Expected outcome: Only the SCADA server can talk to the PLC; all other traffic is logged and blocked. At the device layer, every PLC, RTU, and gateway must have its default credentials changed and remote access limited to a dedicated VPN with MFA. Who: Operations manager with support from IT. What: Change default credentials on every OT device. Use a password manager to store and distribute these. When: During commissioning or during the next planned maintenance window. Expected outcome: Stolen credentials from a generic device cannot be reused on the PLC. At the process layer, implement a change management process for all writable OT parameters. This is the step most operations skip. Who: SCADA engineer. What: Configure the HMI to log every write to a critical parameter (e.g., setpoint, speed, valve position) and alert the operations team when the change does not match a pre-approved maintenance schedule. When: Immediately after network segmentation. Expected outcome: An unauthorized temperature change generates an alert within 60 seconds, not 127 hours.

When an incident hits, the first 60 minutes determine whether loss is limited to a batch of lettuce or extends to a complete facility shutdown. The CISO owns the decision to isolate the OT network from IT. This is not a technical question — it is a risk acceptance question, because isolation may stop the spread but also stop production. The CISO must have a pre-approved playbook that specifies the criteria for isolation (e.g., confirmed PLC manipulation or active lateral movement). The IR team handles forensic collection, but they must work with the SCADA engineer to safely capture PLC memory without triggering a fault. Standard response tools (e.g., winPEAS, Velociraptor) are rarely installed on OT hosts. The IR team must have a dedicated OT incident response kit that includes a ruggedized laptop with offline documentation, serial-to-USB adapters, and a known-clean copy of the PLC firmware for restoration. The handoff that most often stalls is between IR and operations: the SCADA engineer knows the process but not the threat model, and the IR team knows the threat model but not the process. The CISO should appoint a single incident commander who understands both domains — or at least has the authority to force decisions. Legal and comms get involved when there is evidence of data exfiltration or spoilage of product that could harm consumers. One actionable step: prepare a communications template that includes a statement about food safety, not just IT security. Customers and regulators care whether the product is safe, not whether you got hacked.

Here is the hard-learned insight: the most dangerous vulnerability in agrifood is not a missing patch. It is the assumption that because your control system is isolated, it does not need the same monitoring rigor as your corporate network. Time and again, we find that operators who say 'our OT is air-gapped' have the weakest detection because they never look at the logs. Air gaps are rarely real. They are theoretical. The real perimeter includes the laptop the vendor used three years ago, the sensor gateway with cellular backup, and the maintenance engineer's personal phone connected to the WiFi. Treat your OT network as if it is already compromised, because in many environments, the only reason it has not been is that nobody has bothered to look at the Modbus registers. If you cannot detect a write to the temperature setpoint within 60 seconds, you are not secure — you are lucky.

Three lessons generalize beyond this specific threat. First, safety and security converge in industrial environments. A safety system (like a high-temp alarm) that is not monitored for tampering is just a list of events. The same PLC that protects your product can be turned against it. Second, compliance frameworks like NIST SP 800-82 or IEC 62443 provide structure but they do not enforce the specific controls that matter in agrifood — like default credential remediation or out-of-bounds alerting on process values. You need to extend these frameworks with operational context. Third, the attackers do not always want money. The Lyon incident was a state-linked group testing destruction-of-property capabilities. The cost of the attack was not the ransom — there was none. The cost was the product loss, the brand damage, and the operational downtime. If you design your defense only around ransomware, you are leaving half the threat surface exposed.

By mid-2028, I expect to see the first automated attack that combines reconnaissance of exposed ICS protocols with a generative AI agent that can craft a targeted write command based on the specific PLC model. The enabling conditions are already present: large language models can reverse-engineer PLC documentation, and the number of internet-exposed industrial devices continues to rise. Most practitioners will disagree with this, arguing that AI is not reliable enough for control system sabotage. But reliability is not the bar — the ability to cause a one-time failure that destroys a single batch is. By 2029, at least three major agrifood facility operators will be forced to disclose a cyberattack that resulted in a food safety recall. The regulatory pressure will then force mandatory OT incident reporting, which currently does not exist in most jurisdictions. And here is the prediction most will disagree with: the biggest risk is not from sophisticated nation-states but from disgruntled employees or ex-employees who understand the process. A single person with knowledge of the control logic and network credentials can do more damage than any remote attacker. The industry will not address this until the first insider-caused food safety incident makes national headlines.