Unauthorized use of artificial intelligence is becoming one of the main security threats facing businesses, including those in the energy sector. Reports show that as many as 47 percent of users rely on AI tools outside the control of IT departments, increasing the risk of data leaks and cyberattacks. In critical infrastructure industries, the consequences of such practices can have systemic implications.
“AI is a technology that is rapidly gaining ground in Polish companies and beyond. If it is deployed without a well-thought-out strategy and proper governance, it becomes an open invitation to attackers. This technology is in widespread use, which means that almost everyone has access to it today. And because it is so widely available, it can also be used by individuals or organizations that want to harm our business,” Jarosław Zarychta, Head of Business Development at Billennium, said in an interview with Newseria.
The phenomenon known as shadow AI refers to the use of artificial intelligence tools outside the control of IT and cybersecurity departments, most often through employees’ private accounts. According to Netskope’s Cloud and Threat Report: Shadow AI and Agentic AI 2025, the scale of this phenomenon is growing alongside the spread of generative AI. Nearly half of users rely on it in an unmanaged way, while the average organization records around 223 incidents per month involving data being sent to AI applications. The number of users of generative AI tools has tripled year on year, while the number of prompts directed to such systems has increased sixfold.
“In many organizations, employees today have relatively unrestricted access to large language models such as ChatGPT, Google Gemini, or Anthropic solutions, and they use them to formulate queries, or prompts. The problem is that this often happens without oversight from cybersecurity teams and IT departments. The lack of such control creates a real risk for the enterprise, especially in critical infrastructure organizations such as energy companies, because it opens up new attack vectors originating from AI-based tools,” Jarosław Zarychta explains.
According to Microsoft’s analyses, as many as 71 percent of employees have used AI tools outside the official solutions provided by their employer, and some have deliberately bypassed existing security policies.
The Netskope report indicates that the number of available generative AI applications has already exceeded 1,500, while organizations often lack the tools to monitor their use or control data flows. The problem is exacerbated by the fact that employees use many different tools at the same time. The average company relies on at least a dozen different AI applications, which significantly complicates risk management. As a result, invisible data-processing workflows emerge within organizations, along with new attack vectors, including those related to the use of generative AI for content manipulation and impersonation.
“Imagine a situation in which the director of an energy company receives a phone call from the CEO of a major energy group asking them to instruct the accounting department to make a transfer of a specific amount to a contractor, while also saying not to ask questions because the matter is very urgent. That can be generated by artificial intelligence. It is a classic deepfake. If we, as an energy company, do not implement mechanisms to prevent this type of attack, we become vulnerable to, among other things, impersonation of key people in our organizational hierarchy,” explains Billennium’s Head of Business Development.
The development of AI is changing the nature of threats. Generative technologies can be used to automate phishing, create malware, or generate highly convincing deepfake materials that make attacks harder to identify. At the same time, users entering operational, financial, or technical data into AI tools may unknowingly transfer sensitive information to external systems.
The growing importance of security in this area is reflected in EU regulations. Energy companies must now simultaneously respond to the requirements of the NIS2 Directive, national cybersecurity legislation concerning the national cybersecurity system, and the AI Act. This means that the use of artificial intelligence is no longer merely a matter of innovation. It is becoming an element of operational and managerial responsibility directly linked to business and regulatory risk.
“There is a lot of criticism of regulation, but the energy sector has been working with market regulators for years, just like the financial and telecommunications sectors. In this case, the purpose of regulation is to protect consumers and businesses from the negative consequences of using artificial intelligence. Today, the framework for building secure AI within an organization begins with reviewing internal policies and aligning them with procedures and regulations such as those of the European Union,” says Jarosław Zarychta.
The NIS2 Directive imposes obligations on critical infrastructure companies in the areas of risk management, incident reporting, and supply chain security, while the AI Act introduces a framework for the safe use of artificial intelligence. These regulations require, among other things, the identification of AI systems, risk assessments, and the implementation of control mechanisms.
“As a European market, we should build solutions that will protect us effectively against various types of threats. The elements related to this, such as the regulations and requirements contained in NIS2, allow us to prepare better for this new environment. First, we must ensure full control over data, and second, we must ensure that this data is secure from a cybersecurity standpoint, including encryption of data at rest and in transit,” says Mariusz Aksamit, Head of Cloud Delivery Domain at Billennium.
One of the key directions of development is the concept of the sovereign cloud, meaning solutions that ensure full control over data and its processing within a specific jurisdiction. Combined with multicloud architecture, this helps reduce the risk of dependence on a single provider and increases system resilience to disruptions and attacks.
The security of the entire AI supply chain is also becoming increasingly important, from model providers and cloud services to APIs and integrators, as well as the input and contextual data used by the organization. A weakly secured element within this ecosystem can transfer risk to the entire organization, which is particularly significant in the energy sector as part of critical infrastructure.
“This will allow us to make broader use of upcoming AI solutions or new components that will soon appear on the market. Of course, this is also of enormous importance in the context of national security, because the energy sector is a critical sector,” explains Mariusz Aksamit.
The European Union plans to allocate hundreds of billions of euros to strengthening infrastructure resilience, including in the energy and digital sectors. This is intended to support the development of secure technologies and reduce risks associated with cyber threats.
“The energy sector is already trying to implement AI solutions in about one-third of cases, but whether it is truly prepared depends on how it approaches the issue. According to SIG research, 72 percent of energy-related organizations are implementing AI below the security standards we normally expect in IT. So there is still a lot of room for improvement,” assesses the Head of Cloud Delivery Domain at Billennium.
A key element of safe AI implementation is governance, meaning a set of principles, procedures, and tools that make it possible to control how this technology is used within an organization. In practice, this means building a coherent AI management system that includes, among other things, a register of models and use cases, access control and regulatory compliance, monitoring of tool usage, and a full audit trail of user and model activity. Increasingly, this approach also includes cost control for the use of large language models and limiting shadow AI through centralized policies and tools. Billennium offers exactly such a solution in its portfolio: AI Governance Suite, which operates like an “operating system” designed to ensure the secure implementation of AI across an organization.
“For example, lawmakers impose several requirements that are directly linked to AI governance, including the need to verify the AI supply chain end to end, the obligation to report AI-related incidents within 24 hours, and the requirement to conduct regular testing of AI solutions. Because we at Billennium work in this market and have extensive experience, we know that many of these aspects are a major challenge for companies, including those in the energy sector,” says Mariusz Aksamit.
