New Energy World™
New Energy World™ embraces the whole energy industry as it connects and converges to address the decarbonisation challenge. It covers progress being made across the industry, from the dynamics under way to reduce emissions in oil and gas, through improvements to the efficiency of energy conversion and use, to cutting-edge initiatives in renewable and low-carbon technologies.
Exploring AI for energy, starting with sovereignty
17/11/2026
8 min read
Feature
The Energy Institute’s new Applied AI for Energy Committee will bring together energy companies, technology providers and AI specialists to tackle common, non-competitive technical challenges that cannot easily be solved by individual organisations working alone. The Committee draws members drawn from across the energy and technology sectors. The programme will focus on the practical challenges holding back wider adoption of AI and will build directly on the findings of the Sovereign AI Forum, which identified evaluation and procurement, competence and accountability, data rights and operational assurance as priorities for further work. Summarised below is a recent report of the event, which was held under the Chatham House Rule, so comments remain unattributed, writes Stuart King, Technical Manager – Human Performance and Power Systems, Energy Institute.
Key takeaways
- The key issue with AI sovereignty is that organisations need to understand which parts of the AI system they control, which dependencies they accept, and whether they can change those choices later.
- When using AI, the most important capability is being able to test an output, understand the evidence behind it, recognise when the system is unsuitable for a task, and know when human judgement should override the result.
- When considering what constitutes a resilient system, the scale of protection is seen as less useful than how quickly an organisation can identify new risks, understand which systems and assets matter most, and take action when a threat is identified.
- It can be difficult to value preventive systems for which success never occurs, but failures can. Organisations may record the cost of an incident but not the value of the incident that was avoided. A sector-specific value methodology should make avoided loss explicit and should not assume that every benefit can be reduced to headcount or transaction value.
First things first. The forum did not settle on a single definition of ‘sovereign AI’. It is not merely about ownership or control of AI. The most useful working position was that organisations need to understand which parts of the AI system they control, which dependencies they accept, and whether they can change those choices later. The required degree of control differs by workload. A system used for routine content generation does not warrant the same treatment as a system connected to operational infrastructure.
The forum cautioned against holding AI to an absolute sovereignty standard that is not applied to the wider energy system. Energy operations already depend on external equipment, software, services and supply relationships, and critical engineering work has for many years been conducted on simulation software of US origin without anyone raising its provenance. What has altered is the geopolitical environment rather than the software.
The discussion therefore did not treat dependency itself as evidence of insecurity. The practical issue is whether a dependency is understood, proportionate to the workload, capable of being assured, and recoverable if access or supply changes. This qualification is important because it prevents the sovereignty discussion from becoming a case for domestic ownership of every layer.
