Industrial Control Systems Remain Vulnerable to Hacking

Image courtesy of UN Women Asia and the Pacific under Attribution-NonCommercial-NoDerivs 2.0 Generic Deed, resized to 700 x 391 pixels.
The rapid expansion of artificial intelligence is exponentially complicating all things related to cybersecurity, including as it relates to industrial control systems (ICS).
A recent Censys report found that hundreds of thousands of internet-connected devices are running AI products, and that many are likely exposed to attackers. And the risk is increasing rapidly – exposure increased sharply between late 2025 and mid-2026.
This is scary stuff, and has obvious emergency preparedness implications for utilities, so I thought I’d highlight some on the key findings of the report for today’s blog post.
Why Industrial Control Systems Are a Risk
Censys identified about 268,000 vulnerable devices running 44 different AI products, from workflow automation platforms to assistants to large-language-model gateways. The concern is not limited to a vulnerable chatbot or model: AI services depend on layers of supporting software and infrastructure, each of which can become an entry point for hackers.
The lesson for utilities is to inventory and secure the full AI stack, not just the application users see. Simply put, any internet-facing services should be minimized, patched promptly, and protected with strong access controls.
The report highlights the ongoing vulnerability of industrial control systems, which operate equipment and processes in critical sectors. Censys observed roughly 134,000 internet-accessible ICS devices, and more than 80% also ran at least one conventional IT service, such as a web or file-transfer server. Attackers may be able to exploit those familiar services.
Visibility and maintenance are additional problems. Around 70% of exposed ICS devices were on mobile or consumer networks, where operators may be harder to identify and reach. Remote equipment using cellular connections may also miss regular updates. In some countries, a large share of exposed devices sits on a single carrier’s network, creating a concentration risk: one network can offer attackers a more convenient map of potential targets.
Together, the findings point to a practical priority for utilities: reduce unnecessary internet exposure, monitor the software and networks supporting critical systems, and establish clear ownership for patching, even when devices are remote or managed by third parties. And of course, make sure industrial control systems are as secure as possible.
