CISA issues new guidance for secure AI deployment in critical operational technology (OT) systems

Critical infrastructure operators across the West were issued new guidance on Wednesday on how to securely integrate artificial intelligence into operational technology (OT) – all to help reduce the risk of targeted attacks.

The US Cybersecurity and Infrastructure Security Agency (CISA) partnered up with several Western nations to release the new framework, including Australia, Canada, Germany, the Netherlands, New Zealand, and the UK.

The new 21-page Principles for the Secure Integration of Artificial Intelligence in Operational Technology guide outlines four key principles to guide the sector and eliminate security vulnerabilities that, if exploited by threat actors, could pose grave dangers to the populations they serve.

Floris Dankaart, Lead Detection and Response Product Manager at NCC Group, says the coming together of international cybersecurity agencies to address this shared challenge is "rare" and signals the importance of this issue.

Equally important, Dankaart tells Cybernews that “most AI-guidance addresses IT, not OT, meaning the systems that keep power grids, water treatment, and industrial processes running.”

“It’s refreshing and necessary to see regulators acknowledge OT-specific risks and provide actionable principles for integrating AI safely in these environments,” Dankaart says.

Operational technology is defined as a combination of hardware and software systems used to monitor and control physical devices and industrial processes in real time, as opposed to IT, which focuses on manipulating data.

Used in the energy, manufacturing, transportation sectors, and more, prime examples of OT include SCADA and Industrial Control Systems (ICS), as well as the physical technologies involved, such as sensors and valves.

Note that this guide was designed around the “widely accepted” Perdue Model Framework for “understanding the hierarchical relationships between OT and IT devices and networks.”

The four guiding principles

AI integration with critical infrastructure environments is expected to help owners and operators “increase efficiency and productivity, enhance decision-making, save costs, and improve customer experience,” the document states.

Unfortunately, that integration also comes with a myriad of “significant” security risks.

For example, a gradual drift in OT process models or deliberate safety-process bypasses can ultimately impact the availability and reliability of the essential public services the technology manages.

The four principles are outlined below, and broken down even further in the guide.

1. Understanding unique risks and potential impacts of AI
2. Considering AI Use in the OT Domain
3. Establishing AI Governance and Assurance Frameworks
4. Embedding Safety and Security Practices Into AI and AI-Enabled OT Systems

To help operators and owners understand AI, the guide lays out over half a dozen examples of operational technology risks, their potential impact on the entire system, and how to mitigate.

Along with the two risks mentioned above (AI model drift and process bypasses), the first principle lists many others, including challenges to interoperability & complexity, operator cognitive load & false alarms, and AI reliability & hallucinations.

Risk impact stresses everything from productivity loss, reduced safety, and system unavailability, while solutions to vulnerability risk highlight the need to educate personnel and secure the AI development lifecycle.

The second principle, AI use in the OT domain, follows suit by considering various business cases, including everything from examining vendor transparency, push-based architectures, contractual agreements, and predictive maintenance.

Establishing an AI governance structure, such as determining stakeholders, policy and procedures, standardization, testing, and how to embed AI within regulatory frameworks, round out the third principle.

And finally, principle four focuses on Monitoring and Oversight, such as maintaining AI inventory, developing incident response plans, anomaly detection & behavioral analytics, and employing proper failsafe mechanisms.

Possible challenges ahead

Dankaart points out several challenges operators may face implementing the best practices, such as “addressing skill gaps in OT teams, especially where it relates to AI.”

He also says that OT environments are typically much more structured and deterministic than IT environments, which may be at odds with many modern (LLM-based) AI applications.

However, Dankaart also notes that “anomaly detection based on machine learning models has been commonplace in OT threat detection and monitoring for some time and remains a key component of the defender’s arsenal.”

“Balancing these factors and getting down to 'what we really mean' by AI will be key for critical infrastructure owners. Luckily, some of the best practices in OT and AI use overlap; the idea that you must always have a manual fallback procedure, the ability to operate “in island mode” and human-in-the-loop controls, to name a few,” he adds.

The guide further states that for critical infrastructure owners and operators, continuous monitoring, validation, and refinement of AI models is essential for safe and effective OT integration.

Unlock more exclusive Cybernews content on YouTube.