Earlier this year, cyberattackers targeted a Monterrey, Mexico, water utility, after the AI agent Claude helped identify hacking opportunities in the utility’s SCADA and industrial internet of things (IIoT) systems. The move marked a notable shift, because the cybercriminals used AI tools to alert them to operational technology (OT) systems that they may not have been specifically targeting, according to Dragos researchers.

Ransomware attackers have historically infiltrated IT software, but industrial companies and their OT were initially safer from the risk. While IT systems were connected to the internet, OT systems were more closed off. But recent advances in industrial networking have changed that — and granted opportunity to attackers.

Why OT systems are ripe for attack

In ransomware attacks, cybercriminals get access to vital company information housed on its IT or OT systems. The attackers then threatens to make the information public or withhold system access unless the company pays a ransom.

Ransomware attacks on industrial systems continue to rise sharply, thanks to a rise in the use of AI to automate tasks and to also lure employees into parting with credentials, coupled with the industrial sector’s unique vulnerabilities. OT systems can’t be taken offline without immediately affecting productivity, making them valuable targets for attacks.

“These actors want to go after industries that have a low tolerance for downtime and manufacturing and a lot of other industrial sectors fall into that category unfortunately," said Lexie Mooney, senior threat analyst at Dragos and co-author of the company’s ransomware report.

“Stopping a turbine or production line could cause equipment damage or millions in lost revenue,” Mooney said.

Manufacturers are among the worst hit within the industrial sector. In a report that detailed the second quarter of 2026, the industrial cybersecurity firm Dragos identified 1,140 ransomware incidents that affected industrial organizations worldwide, a 12% increase from the 1,020 incidents it recorded in the first quarter of 2026.

The research firm Check Point Research found manufacturing incidents rose 56%, from 937 cases in 2024 to 1,466 cases in 2025. At the same time, the average ransom demand more than doubled from $523,000 in 2024 to nearly $1.2 million in 2025, according to Rebecca Moody, head of data research at the cybersecurity research firm Comparitech.

A cyberattack without the human

AI truly is one of the biggest factors in the present and future of OT cybersecurity.

AI is gaining entry as it evolves from the static, malicious script files that use hardcoded, unchanging routines to automated coding that can scale and adapt to their environments, said Santosh Kumar Jadala, cybersecurity specialist and IT researcher for the University of the Cumberlands, who specializes in the intersection of cybersecurity, organizational governance and human behavior. He is the author of the journal article The Rise of Intelligence Cyber Threats with Artificial Intelligence: Survey of New Attack Techniques.

“Historically, launching an advanced cyberattack required skilled human hackers to manually scan a network, analyze firewall responses, and decide their next step over several weeks. AI automates that operational loop,” Jadala said.

“Threat actors can use AI to automatically probe network perimeters, analyze open ports and test exploit paths in seconds.”

Furthermore, criminals use AI to generate code that alters its structure or signature each time it replicates, so traditional antivirus scanners fail to recognize it. Rather than an attacker having to manually test malware variations, machine learning models can evade detection before the attack is even launched, Jadala added.

Putting the human in the AI cyberattack equation

But AI ransomware hasn’t taken humans entirely out of the system. In an ironic juxtaposition, even as cyberattacks become more automated, attackers continue to rely on social engineering methods to get employees inside the industrial organization to inadvertently leak confidential information, Jadala said. And industrial employees aren’t as likely to suspect ransomware attacks because they haven’t traditionally been targeted.

“New, hyper-targeted, grammatically authentic spear phishing at massive scale is new to the industrial sector,” he said. “This is where AI and natural language processing have most noticeably lowered the technical barrier to entry.”

In other words, when someone, even someone seemingly within an organization, asks for private information, don’t give it up without double or triple checking the source, Jadala said.

“Historically, plant managers, control room operators, and shop-floor engineers at organizations weren’t as heavily focused on the threat of cyberattack as finance or HR personnel,” Jadala said.

“When phishing emails did arrive, they were often easy to spot because of generic greetings, broken grammar, or awkward phrasing.”

AI has eliminated those telltale flaws.

Not only that, today attackers can synthesize publicly available data — such as equipment vendor manuals, maintenance contracts or LinkedIn profiles — to craft flawless, context-aware emails that look like urgent safety bulletins, shift schedule updates or firmware patches from legitimate suppliers, Jadala said.

Even software used to communicate among internal groups is no longer safe, Mooney said.

“Sometimes, someone at victim organization is contacted by the attacker over Microsoft Teams who has come up with a convincing story,” she said.

“They’ll say ‘I’m so and so from IT department’ and they’ll make up some sort of manufacturing issue — maybe an issue with the company’s VPN — and say, ‘Let’s work together to solve this.’”

Often, attackers will use AI to create a domain and a website name that looks exactly like company name and branding, Mooney added.

“Maybe there will be a portal that looks like the company portal and then the employee types in their credentials,” she said. “Legitimate credentials are a gold mine for attackers.”

Also new and concerning is the advent of AI audio cloning, Jadala said.

“An attacker can clone the voice of an operations director or plant manager supposedly placing a quick phone call to request emergency remote access credentials or an urgent password reset,” he said.

The AI cyber wars

One big question is, can AI threat-protection software learn from ransomware threats? That is, do threat and threat-protection software discover how to subvert one another in a continued escalation of AI attack and AI attack cancellation?

Yes, said Jadala, who called this situation adversarial machine learning.

For this, defensive AI models are trained on massive datasets to identify anomalous network behavior or suspicious files. Then, in response, sophisticated threat actors use adversarial AI techniques to gain system entry, Jadala said.

“For example, an attacker can generate small, deliberate perturbations in malware code — changes so subtle that the software still performs its malicious function, but that the defensive AI classifier is fooled into labeling the file as completely benign,” he said. “Defenders then must train their models against those adversarial examples, prompting attackers to devise new evasions. They are continually testing and subverting each other’s algorithms.”

Despite the advent of autonomous ransomware hits, humans are still largely directing the AI tools at present, Jadala added. But he expects human attackers to soon be able to step away from the system.

“The immediate future points toward fully autonomous cyber threats capable of entering a network and independently performing reconnaissance, determining system hierarchies, discovering unpatched firmware, and coordinating multi-node disruption without waiting for human commands,” Jadala said.

“When attacks happen at machine speed, human security analysts will not be able to manually triage logs quickly enough to stop a physical failure,” he added.

The industrial sector will need to implement cybersecurity systems that can automatically detect, isolate, and neutralize cyber threats without human intervention, while treating everyone — even internal employees — as a threat, until proven otherwise, he said.

But Jadala was clear that technology alone will not solve the very real AI ransomware threat to industrial companies. They’ll also need to organize around the AI threats and have clear rules of governance in place, he said.

Meanwhile, Mooney said businesses in the industrial sector need to stay on top of the quickly changing nature of ransomware threats.

“It’s hard for a person to wrap their head around the way the ransomware ecosystem is evolving and the speed it's going,” she said.

Ransomware attacks on OT systems are on the rise and are likely to sharply increase even more. Companies in industrial sectors like manufacturing, as well as utilities or energy, can’t afford to turn their backs to the threat. The fallout from a cyberattack is just too great. OT system owners now more than ever need to install the same kind of firewalls, deterrent, mitigation and remediation software their IT counterparts now use.

About the author

Jean Thilmany is an industrial, manufacturing and engineering writer located in St. Paul, Minnesota. She was an editor at the American Society of Mechanical Engineers for 15 years, where she covered engineering and operating technologies and the intersection of engineering and science. She is now a freelance writer and editor covering similar topics.