The Innovation Platform’s Maddie Hall spoke with Paolo Picca, Innovation Lead at the Office for Nuclear Regulation, about a new regulatory sandbox aimed at exploring the use of artificial intelligence to enhance safety and efficiency in the UK’s nuclear industry.
The Office for Nuclear Regulation (ONR) plays a crucial role in ensuring the safety and security of the UK’s nuclear industry. As an independent regulator, the ONR is responsible for overseeing nuclear safety and safeguarding practices, continuously seeking innovative methods to improve regulatory efficiency. Recently, ONR received funding from the government’s Regulatory Innovation Office (RIO) to implement a regulatory sandbox, which will run until March 2026.
To learn more about this sandbox, the technologies it aims to test, and the potential of artificial intelligence (AI) in nuclear regulation more broadly, The Innovation Platform spoke with Paolo Picca, the Innovation Lead at ONR.
The government funding is promising news for evolving the UK’s regulatory landscape. How was the funding secured, and can you discuss how this project can support the UK government’s ambitions for growth and efficiency through advancements in the regulatory landscape?
In October 2024, the government’s Regulatory Innovation Office (part of the Department for Science, Innovation and Technology) was established to support its commitments to provide investment and commercialise new technologies alongside an increasingly agile regulatory environment.
The opportunity through the RIO’s AI Capability Fund (AICF) offered up to £3.6m of funding to regulators and local authorities to deliver a series of experimental AI projects, aiming to streamline regulation in key growth industries and support the UK government’s ambitions for growth and efficiency.
ONR’s project proposal for AICF originated from discussions with the nuclear industry and other nuclear regulators (including The Environment Agency), addressing a need within the sector to consider opportunities offered by AI within the nuclear sector. With our key project partners, we were very pleased to be successful following a competitive bidding process, as it will complement the work already taking place with the industry.
The ambition of this project is to support discussions across various sectors of the nuclear industry, to help accelerate safe and secure AI deployment beyond the specific examples considered as part of regulatory sandboxing.
We are collaborating with other nuclear regulators, such as the Environment Agency and the Defence Nuclear Safety Regulator, as well as other non-nuclear regulators, such as the Health and Safety Executive.
This collaboration is crucial, as we want to avoid different regulators within the nuclear industry pursuing differing approaches to AI and the desire to share a common way forward.
The overall goal of the project is the safe and secure deployment of AI in nuclear installations and will support the government’s ambitions for efficiency and growth in this sector.
On a broader level, what is the potential for AI use in nuclear regulation? What could ‘smart regulation’ look like?
The UK’s goal-setting regulatory regime, which is technology-neutral and does not seek to prescribe design solutions, already provides a constructive and safe environment within which innovation can thrive.
Underpinned by our enabling regulatory philosophy, we can support industry to capture the benefits of new technology and novel approaches by providing a stable, yet progressive, regulatory environment that enables the delivery of cost-effective safety and security solutions.
As in other sectors, the potential for AI in the nuclear industry is significant, and we are keen to explore various opportunities, understanding the efficiencies that AI solutions can provide, as well as the safety and security implications in their use. From a regulatory standpoint, we take a responsible approach, which supports innovation while still maintaining nuclear safety and security as a priority. In this context, we encourage trials and pilot schemes to build experience before they are adopted as established practices.
We are currently working on various applications of AI in the nuclear sector, from machine learning (ML) to large language models (LLMs). The applications considered for sandbox testing (the practice of giving regulators and industry a safe space to consider potential challenges and solutions) primarily focus on the use of ML as a classifier for computer vision applications, with a machine learning algorithm being trained in a supervised fashion. These applications have been considered for years in other sectors, with broader opportunities now becoming available with the growing power of computers and capacity to manage increasingly larger data sets and increased computational power and data.
We are also running several internal pilot programmes on LLMs to better understand their potential in different applications. One ongoing trial involves using them to help simplify our guidance documents, to make them clearer and more accessible.
Another ONR trial focuses on the use of LLM to support internal knowledge management. Our inspectors possess a wealth of knowledge, and with potential retirements or staff changes, it’s important to capture that expertise for new employees or other colleagues. This also has the potential to help develop programmes for our apprenticeships and graduate schemes.
Additionally, we are experimenting with how LLMs can assist with regulatory assessments. While our inspectors will retain their vital role during assessments and in making regulatory decisions, there is an opportunity for LLMs to help streamline our work by helping them more quickly identify key information within safety case documents.
There are lots of opportunities offered by AI in nuclear power: from using AI to process operational data to support optimisation to the use of AI to drive robotics in hazardous areas. ONR is keen to engage with the UK nuclear industry as it explores options for safe and secure applications, which support the sector’s growth.
Could you elaborate on the two potential test cases mentioned for the AI project and why these were selected? How do you see AI enhancing non-destructive testing and nuclear waste characterisation?
The two test cases were selected following feedback from the industry. Because of the short timescale for this project, we gave priority to test cases where license holders have already explored some of these concepts internally. We were also keen to make sure these test cases were sufficiently broad to attract interest across the nuclear industry, from new build to operating facilities (including nuclear defence) and decommissioning waste management.
One project focuses on waste characterisation management, which presents an opportunity for AI to bring real efficiency. For the examples we are considering, AI can bring improvements to nuclear waste management, providing a more refined classification of intermediate-level/low-level waste with the potential for a significant reduction in long-term storage costs.
On the non-destructive testing side, we are examining opportunities from two perspectives: efficiency and safety. AI use could reduce the time needed for non-destructive testing, which in turn could accelerate new build programmes or operational inspections. AI could also reduce the occupational dose during tests and help with more comprehensive quality checking.
The regulatory sandbox runs until March 2026. Can you explain the concept of a regulatory sandbox and how the previous AI regulatory sandbox pilot influenced this project?
A regulatory sandbox is a safe space for regulators, licensees, and technical experts to collaborate on solutions well ahead of formal assessment and permissioning activities.
It allows for an open dialogue between various stakeholders to better explore the potential benefits and challenges associated with a solution. By doing this, we aim to reduce project uncertainty for innovative solutions under consideration and accelerate their deployment. This also represents an opportunity for the regulator to learn about new approaches and technologies at the early stage of development, helping upskill our staff.
A groundbreaking and world-first regulatory sandbox in the nuclear sector was conducted by ONR and the Environment Agency in 2022-2023, and the results have generated considerable interest among overseas regulators and agencies.
As part of this RIO-funded project, one of our goals is to develop a handbook that outlines how to run a regulatory sandbox and clarify what licensees can expect when engaging in future sandboxing with us.
On a similar note, can you share any details of the project’s progress so far?
Following the confirmation of funding at the beginning of September, we sent out expressions of interest to potential stakeholders, aiming to broaden our network. A launch event was held last October, which was attended by more than 70 people. The next step was an AI workshop in early November to provide a baseline of knowledge on AI design and assurance to support effective discussions.
In parallel, we are progressing the two previously mentioned test cases (waste characterisation and non-destructive testing). These workstreams aim to identify and clarify the context of the challenge and the opportunity we are trying to address as part of this project.
Looking ahead, there will be workshops for both test cases in December, followed by deep dives into each project in January and February. While we don’t currently face any major technical issues, the timeline needs to remain on track as we aim to finalise discussions on both test cases and publish a summary report in the spring of 2026.
Are there any concerns or challenges regarding the wider implementation of AI in nuclear regulation? What does the ONR do to address the concerns, and what learning has to occur before this is implemented on a wider scale?
We believe the UK nuclear industry is generally taking a responsible approach to exploring opportunities offered by AI. This involves engaging with other sectors to learn from their experience, setting up demonstrators to test the technology in a controlled environment, supporting the development of sector-specific guidance and standards where appropriate, and holding discussions at early stages with nuclear regulators.
We believe this approach ensures a licensee continues to meet its legal obligation of reducing the risk as low as reasonably practicable (ALARP) when considering AI applications.
We are also engaging with international peers and have recently developed a pioneer position paper with the US and Canadian nuclear regulators on the regulation of AI (New paper shares international principles for regulating AI in the nuclear sector | Office for Nuclear Regulation), with a follow-up paper being planned during the next year. ONR is also involved in activities at the International Atomic Energy Agency to support the development of a consensus for guiding principles to support a safe and secure application of AI in nuclear power.
In the press release, you mention the importance of mindset changes in fostering innovation. What specific mindset shifts do you believe are necessary within the nuclear industry to facilitate the successful adoption of AI technologies?
Recent discussions at the ONR’s headquarters in Bootle with Heads of Innovation across the UK nuclear industry revealed a consensus that mindset is one of the key barriers to innovation. Our aim is to support changing that mindset by being open to discussions and altering the perception that a regulator would be resistant to an innovative solution or idea.
In this context, ONR is engaging with UK dutyholders as well as at key industry forums like the Safety Director Forum and the Nuclear Engineering Director Forum.
Often, regulators are more open to innovative ideas than organisations anticipate. While this doesn’t mean we will support every solution, our aim is to clarify the reasons behind any potential regulatory challenge to innovative solutions. Clearly communicating the rationale as to why a solution is not suitable for application is a key starting point to enable the industry to innovate further and meet the expectations.
We also believe there is an opportunity to support our staff to adopt a pro-innovation approach by empowering our colleagues and inspectors to engage more confidently with innovative ideas through our innovation ‘products.’ We also support discussions on innovation with weekly drop-in sessions and monthly innovation cafes, inviting internal and external speakers to reflect on opportunities offered by innovation and share learnings.
How does international collaboration influence the development and implementation of AI regulatory frameworks in the UK?
We believe international collaboration is key to ensuring we learn from peer regulators and also help shape consensus supporting a safe and secure deployment of AI in the nuclear sector. We work closely with international agencies on AI, including the NEA and IAEA, and maintain strong links with other leading nuclear regulators in Canada and the US.
ONR was part of the first International NEA’s RegLab (international regulatory sandboxing) in Canada. This event brought together regulators, operators, and technology experts from various countries (including the UK, the US, Canada, Spain, Japan, and South Korea).
This represented an excellent opportunity to discuss examples of applications of AI in enhancing operational efficiencies within nuclear facilities, understanding opportunities explored in other regulatory contexts, and identifying areas of focus for the UK nuclear industry and regulation. A public report detailing the findings and outcomes of this event will be published in the forthcoming months.
ONR is also part of the steering committee of a wider network of innovation at the IAEA (Home), which will enable us to be at the forefront of innovation in nuclear, supporting our horizon scanning on innovation and more general synergies on areas of common interest across various regulatory regimes.
What are the next steps after the project’s end in March? How do you envision nuclear regulation to evolve over the next decade?
Our proposal to the Regulatory Innovation Office outlined our ambitions for the long-term impacts of the two selected test projects. As part of this, we are consolidating our experience in supporting regulatory sandboxing by developing a handbook to inform future activities. We are considering wider learning to create a framework that can reduce regulatory uncertainty on innovative solutions, including AI.
Our wider aspiration is for this project to encourage more UK nuclear licensees to engage with us on innovative solutions to support safe and secure nuclear operations.
With the Environment Agency, we have created various innovation ‘products’ for UK dutyholders to engage with us, from innovation cafes and joint horizon scanning to innovation advice, expert panels, and regulatory sandboxing.
We believe that positive discussions in a safe environment can significantly reduce project uncertainty and accelerate safe and secure deployment of innovation.
Our message is clear – ONR is open to innovation and committed to enabling the safe and secure deployment of AI and other innovative technologies within the nuclear sector.
Please note, this article will also appear in the 24th edition of our quarterly publication.
