Why IT and OT separation depends on controlled access

IT–OT separation requires controlled access

Critical-infrastructure operators need to control access between business and operational systems because maintenance and management can create connections attackers can exploit. Information technology (IT) covers digital systems used to manage information and business activity, while operational technology (OT) monitors or controls physical operations. Separating them means deciding who can cross the boundary, with which permissions and through which connections. Security, engineering and operations leaders share those decisions, including at smaller facilities.

NCC Group’s findings show how access decisions can affect operational continuity. The cybersecurity company sells security services, so it has a commercial interest in organisations investing in protection; readers should weigh its findings and recommendations with that stake in mind. Its ransomware figures describe the wider threat environment, while a small UK power generation facility’s shutdown shows the consequences an individual operator can face.

Ransomware rose over the quarter despite September’s decline

NCC Group recorded a 23% increase in ransomware attacks in the third quarter compared with the previous quarter. Industrial businesses experienced the largest volume, with more than 838 attacks, followed by consumer discretionary businesses with 526. That volume shows the scale of reported activity against industrial operators, although the industrial sector and critical infrastructure are different categories.

Several threat groups accounted for large shares of the quarterly activity recorded by NCC Group. Qilin was the most active, with 365 attacks, or 13% of incidents. The Gentlemen followed with 359 attacks and a 12.5% share; its attack count increased 51% from the second quarter. So the quarterly rise included substantial growth by a group already responsible for a significant share of incidents.

Yet activity fell in September, when NCC Group recorded 821 ransomware incidents, a 26% decline from August. Industrials still experienced 230 attacks, while consumer discretionary businesses experienced 161. North America was the most targeted region that month, with 390 attacks, followed by Europe with 211.

The monthly decline and quarterly increase cover different periods, so they can occur together. NCC Group interpreted the sector pattern as growing pressure on critical infrastructure, including transport, communications, financial services and health. Despite September’s decline, it remained concerned about large organisations and essential-system operators. A disruption at a smaller power facility shows what that pressure can mean for physical operations.

A four-day shutdown shows the operational consequences

A cyber attack on a small UK power generation facility in August forced operations to stop for about four days, according to NCC Group. The interruption reached into power generation, where shutdown duration directly affects continuity. It also shows why smaller operators need to consider their exposure to attacks on critical infrastructure.

Matt Hull, vice president of cyber intelligence and response at NCC Group, explained the significance: “Not only was this attack a reminder that smaller critical infrastructure operators can’t discount themselves as targets, it’s a direct example of threat actors crossing the boundaries between the digital and physical worlds.” His assessment links the facility’s size to the need to plan for physical disruption. Smaller operators still need to account for both the possibility of an attack and its operational effects.

But responsibility remained unresolved: the UK government had not formally confirmed who was responsible, while NCC Group assessed the details as consistent with Iranian-affiliated campaigns against critical infrastructure. NCC Group described attacks on internet-exposed programmable logic controllers, devices that control industrial processes, and attempts to obtain remote access to OT as characteristics of those campaigns. Its assessment described campaign characteristics rather than confirming how attackers reached the facility.

The shutdown shows an operational consequence without establishing a cause for the wider ransomware increase. Hull also said the direct impact of this digital-to-physical capability continued into Q4 in retail. His observation extends the concern to another sector, while the power facility provides a concrete example of a smaller operator losing the ability to carry out physical operations.

Control the access that operations require

To reduce the risk of such disruption, operators need to examine why their IT and OT systems connect. NCC Group linked increasing connectivity partly to third-party maintenance and centralised management, both of which can require access to operational equipment. These arrangements make separation a joint responsibility for security, engineering and operations: teams must understand what a connection is for before deciding how to restrict it.

Even a connection used for authorised work can give an attacker a route into operational systems. NCC Group’s discussion of exposed controllers and attempts to establish remote OT access makes that risk specific. Operators therefore need to decide which connections should exist, who should use them and what those users should be allowed to do. Treating every connection as a configuration error would overlook its operational purpose.

Hull’s recommendations address these decisions through visibility over privileged accounts, network segmentation and secure configuration of remote access. Privileged accounts have elevated authority, so knowing who controls them helps operators identify who can make consequential changes. Network segmentation divides systems into separate network areas and restricts access between them. Secure remote-access configuration governs connections used for work such as maintenance and centralised management.

These controls let operators match access to an approved task and limit what that access permits. An operational purpose alone does not justify retaining every connection, so some systems may need to be isolated or connections removed. Where connections remain, those responsible for continuity need to account for who can use them and what those users can do.

AI introduces two distinct security problems

Access decisions also matter in NCC Group’s AI findings, but the mechanisms differ from those involved in separating IT and OT. One problem comes from attackers using generative AI, which produces outputs from prompts, to assist malicious activity. Another comes from legitimate AI agents, systems that can take actions through connected tools or services, becoming a means of compromise. The first concerns capabilities available to attackers; the second concerns actions taken through access an organisation has granted.

For attacker use, NCC Group described BraZetsu as an AI-enhanced malicious framework for Windows that researchers identified during the third quarter. It later developed into a specialised initial-access toolkit, intended to help attackers gain their first foothold in corporate systems. NCC Group said its operators used models to automate victim profiling, sort and assess data, and rank targets by perceived commercial value. These functions helped attackers assess organisations, identify information worth examining further and prioritise potential victims.

BraZetsu thus illustrates model-assisted targeting alongside the ransomware findings, without explaining the quarterly increase. Legitimate agents create a separate problem when their behaviour or authority becomes unsafe. NCC Group identified misalignment, where an agent’s behaviour diverges from its intended purpose; prompt injection, where malicious instructions manipulate its behaviour; and excessive permissions. Each can allow an accidental or manipulated action to use access the organisation has already granted.

Hull described the pattern: “We‘re witnessing in real time an unintended use case of AI agents, as a pattern emerges of repeated failures rather than isolated incidents. While AI-enabled attacks could be considered a natural consequence of the sophistication of its capabilities, the threat landscape has also been expanded by organisations being compromised by accidental or manipulated agentic actions.” His distinction directs attention to organisations’ deployment and oversight decisions as well as hostile use of AI. An agent’s permissions affect what its unintended actions can cause.

NCC Group cited US government systems and Australian health institutions among those affected in its agent-risk discussion. It used those examples to recommend monitoring how agents interact with systems and data, particularly in critical infrastructure and public services. Such monitoring examines behaviour carried out through granted access, whereas an IT–OT boundary restricts access between environments. Security teams need to assess agent behaviour in its own right, including the actions an agent can take through its permissions and connections.

Key takeaways for decision-makers

• Assign ownership of IT–OT access: Security, engineering and operations leaders share responsibility for connections between business and operational systems. Document each connection’s purpose, authorised users and permissions, including access used by maintenance providers.
• Read ransomware trends across time periods: NCC Group recorded a 23% quarterly increase despite a 26% monthly decline in September. Industrial businesses recorded the largest quarterly attack volume, supporting continued attention to their exposure.
• Plan for physical disruption at smaller facilities: A cyber attack stopped operations at a small UK power generation facility for about four days. Operations leaders need to account for shutdown duration and recovery requirements when assessing cyber risk.
• Match access controls to operational tasks: Security and engineering teams need visibility over privileged accounts, network segmentation and securely configured remote access. Retain connections with an approved operational purpose and restrict what their users can do.
• Assess AI threats through separate mechanisms: Attackers use AI to profile victims and prioritise targets, while legitimate agents introduce risks through manipulated behaviour and excessive permissions. AI deployment owners need to review agent authority and monitor interactions with systems and data.