Managing AI system risk within the energy sector

The assurance imperative

AI assurance will emerge as the foundation for managing AI system risk within energy systems, enabling sector stakeholders to attain digital trust. DNV defines digital trust as the regulator-relevant justified confidence that digital and AI-enabled systems will behave predictably, remain under control and align with public interest outcomes throughout their full lifecycle. Operators must ensure that growing reliance on data-driven and automated decision-making does not introduce new, emergent risks.

AI assurance will play a key role within the context of the UK energy transition. Electrification, decentralization, and the growth of variable renewables are driving greater system complexity and dependence on real-time data and algorithmic coordination. Without appropriate trust frameworks, this shift creates new risks; from system resilience and outage management challenges to potential consumer harm and barriers to innovation. Evidence from network innovation projects shows that a lack of confidence in data quality, model behaviour, and governance is one of the main reasons why digital and AI solutions struggle to move from pilots into business-as-usual deployment.

Technology potential

Effective AI assurance will enable innovation rather than constraining it. By making digital risk explicit, structured, and manageable, assurance allows organizations to experiment with new technologies while avoiding late-stage redesigns, regulatory pushback, or loss of stakeholder confidence. It also provides a common language for aligning network operators, technology providers, regulators, and funders around acceptable risk and evidence requirements, accelerating the path from innovation funding to real consumer and system value.

AI assurance will be critical for future energy systems as AI is increasingly used for forecasting, optimization, fault detection and control-room support. Producers, networks and regulators need confidence not only in performance under test conditions, but in stability, monitoring, oversight, and accountability in live operation. Structured AI assurance is presented as an essential step in moving AI out of innovation programmes and into safety and consumer-relevant decision making.

Making it real

These themes are illustrated within several high-profile energy sector digitalisation projects in the UK. One such project - the Intelligent Gas Grid (IGG) led by British gas network SGN - serves as a practical example of how assurance supports AI-enabled innovation in a safety-critical environment. Assurance, governance, and risk-based confidence building are prerequisites for deploying AI-driven capabilities in real contexts under regulatory scrutiny, and to ensure learnings can be adopted beyond the project itself.

However, AI assurance in the energy sector remains inconsistent. Organisations often provide evidence such as test results, governance approvals, bias checks or monitoring outputs, but do not always connect these into a coherent assurance case. Therefore, Ofgem’s recent consultation on AI assurance in energy is promising. Our view within DNV’s Digital Energy team: we would love to see Ofgem set clearer expectations for what “good enough” evidence looks like for AI systems used in regulated energy activities.

In general, sector regulators can do that by defining minimum assurance expectations for AI use cases that could affect customers, networks, markets or safety. That could cover evidence quality, the need for independence, monitoring, and auditability. Where there is a genuine risk of harm Ofgem, for example, could implement review and enforcement of these controls.

Creating fit-for-purpose AI assurance parameters

Future guidance must be clear on the regulator’s assurance expectations for all stakeholders: operators, vendors and independent assurers/verifiers such as DNV. For example, guidance must clearly address when in the AI lifecycle each assurance activity should happen. Assurance should start during planning and development, not at implementation. This ensures issues get identified earlier, before systems are trialled or relied upon, and reduces the risk of investing in AI that is not feasible, poorly governed or unlikely to justify user confidence.

For safety critical systems, the safety case remains a gold-standard approach to communicate the hazards, risks and controls in place. This is an active area of work for DNV and its academic partners, who include the University of York. DNV also believes organisations should communicate AI assurance in a clear, practical and audience-specific way. Messages should be tailored to what each audience needs to know, using plain language for non-technical groups and more detailed evidence for boards, regulators and assurance teams. The aim should be to explain why the system can be trusted, what has been checked, and what limitations remain.

Achieving Digital Trust

Ultimately the complexities and challenges surrounding the energy transition must be digitally enabled to be resolved, but this enablement will only succeed if it is trusted. Assurance turns AI and digital technologies from mere interesting analytics into trusted infrastructure. By embedding assurance, governance, and lifecycle thinking into how digital systems are developed and deployed, the sector can innovate with confidence, scale safely, and deliver real benefits for consumers and the whole energy system.