In April 1986, a late-night safety test at reactor 4 at the Chernobyl nuclear power plant in what was then the Soviet Union turned into one of the worst industrial disasters in history. The explosion and subsequent fire released radioactive material across much of Europe, displacing communities and leaving a legacy that still shapes public trust in large-scale technological systems. Forty years later, the importance of Chernobyl lies not only in what happened, but also in why it happened.
It is often described as an engineering failure. That’s only partially true. There were known flaws in the reactor’s design, but that alone does not explain the series of decisions that led to the disaster. What Chernobyl ultimately revealed was a deeper vulnerability: the inability of governance structures to manage complex systems under pressure.
The tests conducted that night were aimed at assessing whether the reactor could be safely powered down while maintaining sufficient energy output. As a rule, it was a routine safety drill. In reality, it was already deployed under conditions that significantly deviated from protocol. The Soviet Union’s energy grid in the broader region was under strain. There was pressure to maintain production levels. Testing was postponed and resumed an hour later with reduced staffing and monitoring. Safety systems have been disabled to allow the experiment to continue. Operators working within a rigid hierarchical structure made decisions based not only on technical judgment but also on organizational expectations. These factors were not the only factors that caused the disaster. Together, they created an environment in which failure was not only possible, but likely to fail.
This is the lasting lesson of Chernobyl. Complex systems do not fail only because of technical flaws. When institutional incentives, information flows, and decision-making structures diverge from the realities those systems are meant to govern, they fail.
That lesson resonates today in an entirely different realm. The rapid expansion of artificial intelligence has introduced new forms of infrastructure and new governance challenges and needs. Modern AI systems, especially those operating at scale, require large amounts of computational resources. Data centers currently rank as one of the most energy-intensive forms of infrastructure. In some regions, they are central consumers of electricity, shaping grid investment and long-term planning decisions. This change is occurring rapidly, often outpacing the development of a coherent policy framework.
The risk is not that AI systems resemble nuclear reactors in their failure modes. it’s not. The risks are more subtle. This means that systems supporting AI are being developed and operated within governance structures that are struggling to keep up with their scale and complexity.
Chernobyl shows how such inconsistencies can manifest themselves. In the Soviet case, incentives were defined by production goals and political expectations. Information moved unevenly through hierarchical channels. Local carriers were expected to reconcile competing demands without fully understanding system-wide risks. In such situations, adherence to procedures can be replaced by improvisation, and vigilance can be overridden if necessary.
Today’s AI ecosystems operate under different conditions, but similar tensions are observed. The incentives driving development emphasize speed, capability, and deployment. Competitive commercial and geopolitical pressures drive rapid scale-up. However, the governance of infrastructure, including energy systems, often remains fragmented across jurisdictions and institutions. This raises the familiar challenge of how to ensure that safety, resilience and long-term stability are not subordinated to short-term goals.
One aspect of this challenge is transparency. At Chernobyl, critical information about reactor behavior was not widely shared, limiting operators’ ability to make informed decisions. In modern AI systems, opacity takes a different form. The complexity and unique limitations of machine learning models can make it difficult to assess how a system will behave under stress or at scale. When such systems are integrated into critical infrastructure, the risk of uncertainty increases.
Another aspect is accountability. The Chernobyl disaster exposed a system in which responsibility was dispersed and often obscured. Decisions were shaped by the institutional context, but accountability was assigned ex post. Similar questions arise in the governance of AI. When systems fail or infrastructure becomes strained, it is not always clear who is responsible. Whether it be developers, operators, regulators or the broader policy environment.
Energy policy becomes even more complex. The growth of AI is raising new questions about how power is generated, distributed, and prioritized. Should specific uses of computing power be treated as strategic resources? How should the power grid balance industrial and public needs? What are the mechanisms to ensure that rapid expansion does not exceed the resilience of the system? These are not purely technical questions. These are governance issues that require coordination between public and private actors and even across borders.
A simple counterfactual helps clarify the stakes. If AI had existed in the Soviet Union in 1986, could it have prevented the Chernobyl incident? More sophisticated surveillance and predictive modeling could have identified dangerous conditions sooner. The decision support system could have provided clearer guidance to operators navigating an increasingly volatile situation.
However, such possibilities depend on more than technical abilities. You need an organizational environment where data is trusted and systems are allowed to influence decision-making. In the Soviet context, where the flow of information was often constrained by centralized authority and political considerations, even advanced tools may have struggled to change outcomes. A technology is only as effective as the governance structure within which it operates.
This insight remains important. There is growing interest in using AI to manage complex systems such as energy grids and industrial processes. These applications have great potential. You can increase efficiency, improve predictions, and support more informed decision-making. But they also introduce new dependencies and new forms of risk, especially when introduced at scale without corresponding advances in monitoring and coordination.
The challenge for policymakers is not to slow technological progress, but to shape the conditions in which it unfolds. This requires a shift in emphasis. Governance cannot be treated as a second layer applied after the system is built. Incentives, transparency, and accountability need to be attended to and integrated from the beginning.
Adaptive regulatory frameworks are part of the answer. Static rules can’t keep up with rapidly evolving technology. Instead, governance must be able to learn, adjust, and respond to new information. Organizational capabilities are equally important. Effective supervision depends not only on formal authority but also on expertise, resources, and the ability to deal with technical complexity.
Perhaps most importantly, energy policy and technology policy need to be considered together. The expansion of AI is based on physical infrastructure and has significant implications for resource allocation and long-term planning. Treating these domains in isolation risks overlooking systematic interactions that can lead to instability.
Chernobyl does not provide a simple blueprint for the present. The world has changed and so has technology. But it is a stark reminder of how complex systems behave when governance is poor. Failures are rarely caused by a single error. They result from the accumulation of small deviations and are made worse by pressure.
Forty years later, the most important question is not whether we have learned the technical lessons of Chernobyl. It’s whether we have absorbed that institutional stuff.
*Ioannis SidiropoulosLegal Consultant, LL.M, MA, PhD (c) AI Governance and Sovereignty
