At the Portfolio AI in Energy 2026 conference panel, experts agreed that artificial intelligence is no longer just a future prospect for the energy sector but is being deployed across stages from project development to operations. They warned, however, that reliable and beneficial AI deployment requires proper measurement, data use, auditability and cybersecurity frameworks.
Main use cases and examples
- Csonth Dániel, founder and CEO of Veridue AI, said AI is particularly valuable in project development and transaction preparation: information locked in contracts and other unstructured documents can now be processed, potentially shortening due diligence from weeks or months to hours or days.
- Hajós Balázs of Schneider Electric said AI appears both in the company’s products and services and in transforming internal operations. Schneider already uses industrial “copilot” features that let operators interact with complex software in natural language; such tools can increase an operator’s day-to-day efficiency by up to 90–95 percent for some tasks.
- In district heating, an AI system that optimizes outgoing water temperature and forecasts heat demand can reveal energy-efficiency reserves, provided IT and OT systems are strictly separated and human oversight is maintained.
Why many industrial projects stall: auditability and controlled failure
Farkas Péter, founder of ArchonLayer, noted that many industrial AI projects never reach production because they lack auditability and mechanisms for controlled failure. Companies must be able to demonstrate that an AI-driven process will not cause financial loss, physical risk, operational disruption or regulatory issues; ArchonLayer develops solutions that log, supervise and produce interpretable reports about AI decisions.
Model fine-tuning and industrial requirements
Jánki Zoltán, assistant professor at the University of Szeged’s Software Development Department, said open-source or general models can often reach 70–80 percent performance quickly, but industrial settings frequently require accuracy, stability and reliability above 95 percent. Achieving that demands fine-tuning and adaptation to the specific task.
Build in-house or buy external software?
Csonth Dániel suggested internal development makes sense for horizontal, enterprise-wide processes and integration of proprietary systems. For solutions that require deep domain expertise, continuous model updates or benchmarks learned from multiple industry players’ data, external software providers often remain the rational choice.
Data needs and measurement gaps in Hungary
Hajós Balázs emphasized that AI compatibility depends on measurement. In Hungary, relatively little industrial data is collected, and the data that is gathered is underutilized. For example, a high share of data gathered by transmission system operators may not be meaningfully used; analyzing and selectively publishing such data could open opportunities for startups and smaller technology firms.
Regulation and risk management
Jánki Zoltán explained that the European Union’s AI Act should be seen not as a prohibitive framework but as one that promotes conscious, risk-based AI use. Companies must assess which AI applications they use, classify their risk level, and meet associated documentation, transparency and control obligations. Logging decisions and proving system reliability become critical with high-risk AI systems.
Cybersecurity concerns with AI agents
Farkas Péter warned that AI agents now often do more than suggest—they can act—which creates new attack surfaces. Systems should log what decisions are made, in what context, and whether agents have access to resources they should not. He noted that prompt-injection type attacks cannot be fully eliminated due to model architectures, so monitoring communications, detecting anomalies and safely halting processes when necessary are essential defenses.
Future challenges and opportunities
Panelists identified two interlinked trends shaping the coming years:
- Rising energy demand of data centers: Hajós Balázs reminded the audience that large data centers now account for about 1 percent of global energy consumption, and that share could rise to as much as 3 percent by 2030. Managing this growth is a key challenge, and one opportunity is for AI to optimize its own energy use.
- A self-reinforcing cycle: Csonth Dániel said a possible positive feedback loop would be if faster deployment of renewable projects enables more AI infrastructure, which in turn accelerates AI development and further project rollouts.
Other highlighted risks included the potential weakening of cryptographic methods (noted by Farkas Péter) and the growth potential in hierarchical arrangements of AI agents—multi-level expert and executor systems that can automate cooperation (noted by Jánki Zoltán).
Conclusion
The conference panel made clear that AI is already creating value in the energy sector, but realizing that value at scale depends on measurement and data use, auditability, and cybersecurity. In Hungary, limited data collection and utilization are significant constraints, while EU regulation and novel technical risks will shape how AI is adopted going forward.
