Sellafield and Nuclear Decommissioning Authority
Pioneering DANI2, the first AI colleague for nuclear decommissioning
Decommissioning the UK’s first nuclear power station is a mammoth undertaking, driven by safety cases that require expert knowledge management. Retrieving this knowledge enables complex reports to be generated faster; currently, they can take up to a year to issue.
Together with Sellafield, the birthplace of the UK’s nuclear industry – and the UK’s Nuclear Decommissioning Authority (NDA) we partnered to use AI to challenge the status quo and enable the industry to do more for less without compromising on safety.
The result is DANI2 the industry’s first AI Agent, a digital colleague powered by Generative AI to transform knowledge management. With DANI2 in place, Sellafield predicts that it will save £40m over 10 years.
Digitalising nuclear safety documentation
Sellafield was the UK’s first nuclear site and is part of the UK’s Nuclear Decommissioning Agency (NDA). The NDA is responsible for cleaning-up the UK’s highest nuclear risks and hazards, and safeguarding nuclear fuel, materials, and waste. Ensuring its safe maintenance and dismantling requires continuous access to, updates of, and creation of critical documentation, ranging from operating procedures to safety cases.
Many of these records date back over 60 years, often existing only on paper and proving difficult to retrieve. The complexity of the site and the rigorous approval processes mean that authoring and/or updating just one of these documents can potentially take a long time. Additionally, Sellafield faces a major knowledge gap: with institutional knowledge being lost as experienced staff retire.
Sellafield knew that digital and AI solutions would provide huge efficiencies for long-term information storage. It needed a partner to enable this transition from paper-based processes to digital formats to streamline processes for updating, verifying, and approving information. This shift is essential as the volume of data grows and the number of engineers may reduce.
At Sellafield, we’re flipping the script on how the nuclear industry can work. Decommissioning the UK’s first nuclear power station isn’t just a technical challenge, it’s a once-in-a-generation opportunity to rethink how we manage knowledge, make decisions, and keep our people safe. With the use of state of the art technologies such as AI, we’re turning dusty archives into living intelligence. Between 2027 and 2030, I want our teams to see AI not as a tool, but as a trusted digital colleague, helping us move faster, think smarter, and do more with less. It’s bold, it’s exciting, and it’s exactly what the future demands.”
Breaking new ground
Our team of AI, data science, digital engineering, design, regulatory, and nuclear industry experts began by interviewing engineers from eight different business functions, uncovering over 80 AI and GenAI use cases. Our focus was on identifying unmet needs – what are the things that engineers cannot do today that would be transformative for them.
Collaborating as one team with engineers from Sellafield and the NDA, we prioritised five high-impact, high-feasibility use cases to evolve into individual features of DANI2. We designed each feature with the end-user in mind, supported by an interactive front-end built according to UK Government Digital Service principles.
We established a comprehensive validation framework that allowed engineers to assess each response thoroughly. Leveraging an agile methodology, we conducted multiple testing cycles, incorporating feedback at every stage to refine the final solution.
To showcase DANI2’s flexibility, we integrated both OpenAI’s LLM and an open-source, quantised LLM into our infrastructure, demonstrating that the solution could also be deployed on-premises.
Finally, we built in guardrails to ensure DANI2 operates in accordance with Sellafield’s policies. For instance, if a user asks, “How do I make a pizza?”, DANI2 will respond that it is not designed to answer such queries, despite having the knowledge to do so.
Proving AI’s worth
DANI2 took just three months to develop and deploy, and engineers are already seeing results. It currently helps each engineer:
- Talk to thousands of documents in real time, for example, helping safety case engineers find relevant good practice
- Introduce game-changing operational efficiency, for example, detective work that could take up to a year in the past now takes minutes
- Help improve writing quality, for example, by helping engineers check whether they need to provide more substantive evidence against specific requirements set by the regulator
- Automatically check the document has been appropriately referenced
- Analyse Safety Cases and provide recommendations to ensure it complies with MUST-SHOULD-COULD requirements.
We need to invest taxpayers’ money wisely, ensuring we don’t pay for the same solution multiple times. The building blocks we’ve created with DANI2 will unlock tangible economic value across the entire NDA group.”
The success of these initiatives provides a clear roadmap for other highly regulated industries looking to harness AI to tackle complex challenges. Building on this momentum, Sellafield and the NDA now have the confidence to expand their AI capabilities, integrating diverse data sources, and strengthening their workforce. DANI2 will help address critical considerations, such as cybersecurity, organisational change, and mitigating the risks of over-reliance, that are essential for sustainable, large-scale AI adoption.
By embracing new ideas and challenging the status quo, Sellafield and the NDA are setting higher standards in nuclear decommissioning and paving the way for responsible, future-oriented innovations across the sector.
This project has been a resounding success, truly challenging the status quo in how we can embed AI in our approach to knowledge management and decision-making in the nuclear industry. The PA team have done a fantastic job in bringing this to life and helping train our team so that we can be self-sufficient.”
Explore more
