Quick Takeaways
-
AI attack planning is advancing: Frenos tested models across 16,980 OT-security tasks; three of six frontier models produced workable attack plans in roughly 90% of initial attempts.
-
Results remain unreliable: Repeated tests exposed inconsistent performance, missing steps and assumptions about access attackers may not have—meaning skilled human guidance is still important.
-
Autonomous industrial attacks were not demonstrated: Models worked from supplied target information and simulated scenarios; Frenos estimates reliable attack planning could emerge within about two years.
- Defenders should prepare now: Frenos recommends strengthening detection and response, using AI to assess their own environments, and reviewing exposure across entire systems—not just individual vulnerabilities.
What’s the Problem?
Frenos, an operational technology security company, warns that critical infrastructure operators may have about two years to prepare for reliable AI-assisted cyberattack planning. In its inaugural State of AI in OT Security report, the company tested frontier and open-weight models across 16,980 decisions, plans, and simulated runs. Three frontier models produced workable attack plans in roughly 90% of attempts. However, results varied when tests were repeated, and some plans skipped necessary steps or assumed access the model did not have.
The warning concerns operators of industrial systems, including energy infrastructure; the report does not show that AI independently attacked real facilities. Instead, the models worked from information supplied by researchers, and their errors could require an experienced person to identify and correct. Frenos CTO and co-founder Harry Thomas, who conducted the research, said operators should use the estimated two-year window to prepare. CEO Brian Proctor also argued that controlling access to powerful AI models is part of protecting critical infrastructure.
Security Implications
Frenos warns that advances in AI could help attackers plan more convincing, targeted cyberattacks—including attacks on industrial control systems (ICS) that manage equipment and essential processes. This risk is not limited to utilities or factories: any business can be exposed through its networks, suppliers, or service providers. If an attack disrupts operations, the effects can spread quickly: missed orders, lost revenue, unsafe conditions, stolen data, legal costs, and damaged customer trust. Moreover, even a company with no industrial systems may depend on affected infrastructure or vendors. AI does not make an attack inevitable, but it may make preparation faster and more effective, so businesses should treat cybersecurity as an operational priority.
Possible Remediation Steps
Timely remediation limits the window in which AI-assisted attacks can exploit weaknesses and disrupt essential services or endanger safety.
Priority Actions
- Assess exposure: Inventory IT and OT assets, map critical dependencies, and prioritize systems by safety, operational, and service impact, consistent with NIST CSF’s Identify function.
- Harden access: Enforce multifactor authentication, least privilege, strong credential controls, and tightly restricted vendor and remote access.
- Segment networks: Separate operational technology from corporate IT and the internet; limit traffic between zones and disable unnecessary services.
- Patch safely: Prioritize exploitable vulnerabilities and exposed systems. Test updates in a representative environment, schedule controlled deployment, and use compensating safeguards where patching could disrupt operations.
- Improve detection: Monitor OT network traffic, system changes, authentication, and unusual process behavior; establish baselines and ensure alerts reach responders promptly.
- Prepare response: Maintain tested incident procedures for isolating affected systems while preserving safe operations, including clear escalation paths among cybersecurity, engineering, and facility teams.
- Protect recovery: Keep offline or otherwise protected backups of critical configurations and data, and regularly test restoration and manual operating procedures.
- Coordinate and review: Share relevant threat information with trusted sector partners and vendors, exercise AI-enabled attack scenarios, and update risk decisions under NIST CSF’s Govern, Respond, and Recover functions.
Advance Your Cyber Knowledge
Stay informed on the latest Threat Intelligence and Cyberattacks.
Learn more about global cybersecurity standards through the NIST Cybersecurity Framework.
Disclaimer: The information provided may not always be accurate or up to date. Please do your own research, as the cybersecurity landscape evolves rapidly. Intended for secondary references purposes only.
Cyberattacks-V1
