How AI is changing cyber risk across speciality chemical supply chains
Chris Anley, chief scientist at global cyber security firm NCC Group, examines how organisations can adopt AI without introducing vulnerabilities that ripple through critical supply chains
The speciality chemicals industry is easy to overlook when we think about critical infrastructure, because the chemicals themselves are rarely the finished product. Instead, they are essential inputs into something else: a pharmaceutical product, an electronic component, a watertreatment process or another industrial application. That means the resilience of a chemical supply chain can have consequences well beyond the organisation producing the chemical itself.
The chemicals industry is becoming increasingly connected. Manufacturing environments are linked to enterprise IT, suppliers, customers, logistics providers and cloud platforms. Artificial intelligence( AI) is adding another layer of technology and dependency, with organisations exploring its use across areas such as process optimisation, predictive maintenance and logistics. The opportunity is significant, but so is the need to understand what happens when AI becomes part of a complex supply chain.
Accelerating an existing challenge
AI introduces new categories of cyber threat, and it can amplify and accelerate risks that organisations already face. Chemical manufacturers have long had to manage the security of operational technology, enterprise systems, suppliers and third party technology. AI can introduce automation of planning, analysis and decision-making, of IT infrastructure and potentially even of control.
In our consulting work, we are increasingly seeing AI systems deployed as a means of automation and control. The crucial lesson from our attack simulation work is that inputs to AI systems can carry data, but may also carry instructions from attackers, and it is exceedingly difficult to separate the two.
‘ Prompt injection’ and‘ jailbreaks’ are extremely effective attack techniques. Once an attacker is able to issue an instruction to an AI system in this way, everything the AI system can control becomes another tool for the attacker. At that point, the attacker can simply ask for sensitive data, or for an administrative or operational action to be carried out, and the system will obey.
Before AI, the attacker would have to use carefully constructed exploits to increase privilege and move through the network to achieve their objective through a series of complex technical steps. With AI as the mechanism of control, the attacker simply asks the AI for what they want and the AI provides it, promptly and efficiently.
This underlines the importance of traditional, reliable security practices, governance and operating models. An organisation may be able to introduce an AI capability in weeks, but doing so securely, understanding its dependencies, assessing the security of its suppliers and adapting processes and controls to ensure operational resilience, can take considerably longer. The resulting gaps can cause organisational tension and the compromises that emerge from this are where risk can accumulate.
52 SPECIALITY CHEMICALS MAGAZINE ESTABLISHED 1981