The Office for Nuclear Regulation (ONR) has today published its report on regulatory sandboxing and accelerating safe and secure deployment of Artificial Intelligence (AI) in nuclear installations
The seven-month project was completed with funding from the government’s Regulatory Innovation Office (RIO) Artificial Intelligence Capability Fund and focused on AI in computer vision and data classification, using sandboxing to explore how regulation can support the nuclear industry to safely realise the potential benefits of AI.
Regulatory sandboxing brings regulators and industry together in a structured and collaborative environment to explore how assurance can be provided for novel innovations before wider deployment.
ONR and the Environment Agency piloted this approach with AI in 2023, and the lessons learnt informed this latest project.
It examined two specific AI applications relevant to the UK nuclear industry, both using supervised machine learning to analyse and interpret computer vision data, training it to look at images or video footage and identify, categorise or flag the results, with significant potential uses in monitoring, inspection and safety.
ONR led the project in collaboration with the Environment Agency, the Health and Safety Executive, and the Defence Nuclear Safety Regulator, as well as several major industry partners including Rolls-Royce Submarines, NNB Generation Company (Hinkley Point C), Sellafield Limited, and Nuclear Restoration Services Limited.
The project demonstrated our commitment to agile working, bringing together more than 20 people from ten organisations and focusing on four examples from the industry through a series of technical workshops.
It also drew on experience from outside the nuclear sector, with valuable insights from the health and safety, medical and automotive industries helping to broaden thinking and challenge assumptions.
The report’s findings are based on discussions from the sandboxing project but do not represent a formal regulatory position.
Any future decisions about AI systems in the nuclear sector will be made by an independent regulatory team, following ONR and Environment Agency policy, principles and processes.
However, ONR’s findings conclude that the UK's outcome-focused regulatory framework is well-suited to accommodating AI with the safety of a specific application depending on how it is designed and implemented.
Paolo Picca, ONR’s Innovation Lead, said:
“This project shows how we can cooperate across various sectors of the UK nuclear industry to advance our understanding on the opportunities AI can offer.
“The learnings from this project will provide helpful foundations for future engagements on specific applications, to be progressed as part of our standard regulatory engagement processes. ONR remains committed to enabling the safe and responsible deployment of AI in nuclear installations.”
The project also encouraged more ambitious thinking about AI within the nuclear sector. By examining applications beyond those initially proposed by industry, it demonstrated that a wider range of AI uses can be pursued safely within the existing regulatory framework.
Key learning emerged in three areas which ONR will share with the wider industry: the technical skills and competences needed to develop and assess AI systems; how to provide appropriate assurance for AI; and how AI fits within existing nuclear safety cases.
ONR is finalising a handbook to support future sandboxing initiatives, while also investing in further training for inspectors, and exploring broader frameworks for regulatory innovation that could apply to AI and other emerging technologies.
A formal review of the project will take place in a year to assess progress and discuss how use of the two AI applications have developed.
The Regulatory Innovation Office (RIO) is an office within the UK Department for Science, Innovation and Technology
The RIO’s AI Capability Fund is providing up to £3.6 million of funding to a selection of regulators to deliver a series of novel and experimental AI projects. These awards will enable tests of how AI can increase productivity and impact within a variety of different industries.