Ofgem have launched a consultation to discuss how they should run a sandbox for AI development in the energy industry.
A cautious step into an already active landscape
This recent consultation from Ofgem on establishing an AI Technical Sandbox is both timely and, in some respects, overdue. Artificial intelligence is often framed as an emerging capability within the energy sector, but the reality is that AI has already arrived—quietly embedded in operational processes, analytics platforms, and, most visibly to consumers, customer service channels.
Energy suppliers across the UK now routinely deploy AI-driven chatbots as the first point of contact for customers seeking help online. These tools are intended to reduce costs and streamline service, yet many consumers would argue they have achieved the opposite: creating friction, delaying resolution, and making it harder to access a human when it matters most. The consultation speaks at length about innovation, safeguards, and responsible deployment, but it inevitably raises an uncomfortable question. If AI is already being used in ways that materially affect customer outcomes, will the regulator consider whether some applications should be paused, improved, or even withdrawn until a proper testing framework—such as the proposed sandbox—is in place?
This tension between innovation and lived experience sits at the heart of the consultation.
What the consultation is proposing
At its core, the document sets out a proposal to establish a time-limited AI Technical Sandbox: a controlled environment in which companies can test AI applications using realistic data, scenarios, and operational challenges without exposing the live energy system—or customers—to unmanaged risk. The intention is not to approve technologies, but to learn from them. The sandbox would allow participants to explore how AI performs, where regulatory uncertainty exists, and what safeguards may be required before wider adoption. – Effectively, Ofgem want to find out what they don’t know, which I think is an intelligent approach to take.
The pilot is expected to run for 12 months and will focus on structured experimentation rather than open-ended trials. Participation would be led by licensed energy companies and other regulated actors, who may collaborate with technology providers or academic institutions. Ofgem’s emphasis is clearly on ensuring that accountability remains with entities already bound by the existing regulatory framework.
Importantly, the sandbox is framed as a learning mechanism for the regulator as much as for industry. Rather than attempting to draft rules for technologies that are still evolving, Ofgem is signalling a preference for evidence-based regulation: observe first, regulate later.
AI’s promise for the energy transition
The consultation highlights the potential for AI to play a significant role in delivering a smarter, more flexible, and lower-carbon energy system. Use cases range from forecasting demand and managing distributed energy resources to improving network planning and identifying system faults before they occur. These are not abstract benefits. If implemented effectively, such tools could reduce system costs, improve resilience, and support the integration of renewable generation at scale.
From a policy perspective, this aligns closely with the UK’s broader ambition to decarbonise while maintaining affordability and security of supply. AI offers the ability to manage complexity that would be extremely difficult to handle through traditional analytical approaches alone. The electricity system of the future—characterised by millions of small assets, variable generation, and dynamic demand—will almost certainly require some form of automated intelligence to operate efficiently.
Yet technological capability does not automatically translate into public trust.
The gap between innovation and consumer reality
While the consultation understandably focuses on high-value system applications, it is impossible to ignore the area where most people already encounter AI: customer service. Here, the experience has often been poor. Many chatbot systems struggle with nuance, misinterpret straightforward queries, and create circular interactions that frustrate rather than assist. For vulnerable customers or those facing billing issues, this can be more than an inconvenience—it can become a barrier to accessing support.
This raises an important regulatory consideration. If AI is to play a growing role in the sector, should there be clearer expectations around transparency, accessibility, and minimum performance standards? And should certain customer-facing uses be subject to stricter scrutiny than back-office optimisation tools, given their direct impact on households?
The sandbox could provide a venue to answer these questions, but only if it meaningfully engages with real-world applications rather than focusing solely on future innovations.
How the sandbox would work in practice
The proposed model includes defined entry criteria to ensure that projects deliver sector-wide value rather than narrow commercial gain. Participants would need to demonstrate that their use case is genuinely innovative, that it addresses a regulatory uncertainty, and that it can be tested in a measurable way. Oversight would be provided through a steering structure designed to maintain transparency and avoid conflicts of interest.
Another notable feature is that participation during the pilot phase would not involve fees. This reflects a desire to encourage engagement while maintaining neutrality between organisations of different sizes. The regulator also intends to publish insights and lessons learned so that benefits are shared across the industry rather than captured by individual participants.
This emphasis on collective learning suggests that Ofgem sees the sandbox not simply as a technical exercise, but as a foundation for future policy development.
A broader shift toward experimental regulation
The AI Technical Sandbox sits alongside other innovation initiatives already underway within the regulatory landscape. Taken together, they point to a gradual shift away from static rule-making and toward adaptive governance models that evolve alongside technology. In sectors experiencing rapid digital transformation, this “test-and-learn” approach is becoming increasingly common.
For energy, this may be particularly important. Infrastructure investments are long-lived, markets are highly regulated, and mistakes can have systemic consequences. A structured experimentation environment offers a way to reconcile the need for caution with the imperative to modernise.
My perspective—and next steps
As someone working within this space, I welcome the consultation as a necessary and pragmatic step. The industry does need a forum to examine how AI can be deployed responsibly, and regulators are right to seek evidence before reshaping the rules. At the same time, the conversation must remain grounded in the realities customers are already experiencing today. AI in energy is not a distant prospect; it is already shaping interactions, sometimes badly.
It will be interesting to see whether consultation responses push Ofgem to look more closely at current deployments, particularly customer-facing tools, and whether there is any appetite to set clearer expectations while sandbox trials are underway.
I will be completing the consultation and intend to share my response publicly within the next two weeks. There is a valuable opportunity here for industry participants, consumer advocates, and technology providers to help shape a framework that enables innovation without losing sight of service quality and accountability.
A moment to shape the rules before they harden
The consultation represents an inflection point. AI will undoubtedly become embedded across the energy value chain, but the way it is introduced—and the safeguards that accompany it—are still open to influence. By choosing to explore these questions now, Ofgem has created a window for collaborative design rather than retrospective correction.
Whether that window leads to genuinely better outcomes will depend on how widely stakeholders engage, how honestly current shortcomings are acknowledged, and how effectively lessons from the sandbox translate into practical regulation.
I’ll be back once I’ve written my response.
Thanks all,
Richard Winstone
The Regulator Guy.
