Project overview

This SPRITE+ secondment brought together researchers from the University of Manchester and resilience specialists at RINA to explore how AI can strengthen the resilience of critical infrastructure and better protect vulnerable communities during floods, droughts and other major disruptions.

Modern society depends on interconnected systems such as electricity, transport, water, communications and healthcare. When one system fails, the effects can rapidly cascade into others, disrupting essential services and placing vulnerable people—including older adults, people with disabilities and those dependent on medical equipment—at greatest risk.

The project initially set out to investigate whether Large Language Models (LLMs), machine learning and agent-based modelling could enhance existing resilience assessment methods. However, as the work progressed, it became clear that the available historical flood datasets were not sufficiently representative of local conditions in Essex to support credible machine-learning models. Rather than forcing an inappropriate technical solution, the project adapted its approach and focused on a more valuable question: Can AI make it easier to build the high-quality knowledge needed for future resilience modelling?

Given the six-week duration of the secondment, the objective was not to create a complete operational database. Instead, the project explored the feasibility of using LLMs, Google Maps and AI-assisted information retrieval to identify, organise and interpret information from a wide range of publicly available sources. These included Environment Agency flood records, weather and hydrological datasets, UK flood-risk and insurance tools, local authority reports, infrastructure databases, news archives and online mapping services.

The work demonstrated how AI can rapidly transform large volumes of fragmented online information into structured knowledge. Prototype hazard-event templates, representative inventories of critical infrastructure and vulnerable locations, and realistic Essex flood and drought case studies were developed to illustrate how future resilience datasets could be assembled far more efficiently than through traditional manual methods.

A further outcome was a stronger emphasis on people-centred resilience. The project explored how AI could help identify vulnerable communities and understand how infrastructure failures affect people during emergencies. Concepts were developed for linking information on vulnerable populations with hazard scenarios while respecting privacy through federated and privacy-preserving data approaches.

The impact of the secondment extended beyond its original objectives. Several ideas developed during the collaboration—including AI-assisted hazard intelligence, human-centred resilience planning and vulnerability modelling—were incorporated into INTRA-CARE, a Horizon Europe proposal led by RINA. This resulted in the University of Manchester joining a new European consortium working to improve the resilience of healthcare infrastructure and communities across Europe.

Ultimately, the project showed that AI's greatest value may not lie only in making predictions, but in helping researchers and practitioners build the trusted, high-quality knowledge needed for better resilience planning and more informed emergency decision-making.

Objectives

The project aimed to:

1. Explore the use of AI for critical infrastructure resilience, including Large Language Models (LLMs), machine learning and agent-based modelling for hazard analysis and emergency planning.

2. Assess opportunities to enhance RINA's resilience assessment methods, particularly for analysing infrastructure interdependencies and cascading failures.

3. Investigate AI-assisted approaches for developing resilience datasets, using publicly available hazard, infrastructure and community information.

4. Strengthen collaboration between the University of Manchester and RINA, establishing a foundation for future joint research and funding opportunities.

Changes to the original objectives:

The overall objective of applying AI to improve resilience planning remained unchanged, but the technical approach evolved following a review of the available data. The original plan to evaluate advanced machine-learning models using historical European flood datasets was not pursued because the data were considered insufficiently representative of Essex's local hazard characteristics.

Instead, the project focused on assessing the feasibility of using LLMs, Google Maps and AI-assisted information retrieval to collect, organise and structure hazard intelligence from publicly available sources. Given the six-week duration of the secondment, the emphasis was on developing proof-of-concept workflows, data templates and representative case studies, rather than a complete operational dataset.

The project also expanded its scope from infrastructure resilience to human-centred resilience, exploring how AI can support the identification of vulnerable communities and privacy-preserving resilience planning. These concepts subsequently contributed to the development of the Horizon Europe INTRA-CARE proposal, extending the impact of the secondment beyond its original objectives.

Activities

Stage 1 – Project initiation and knowledge exchange

The secondment began with a series of technical meetings with the RINA project team to understand the objectives, methodology and outputs of the MEDiate project, particularly the Essex testbed and its multi-hazard resilience assessment framework. These meetings introduced the available datasets; existing resilience models and the practical challenges faced in modelling flood and infrastructure risks.

Stage 2 – Review of existing machine-learning approaches

Working closely with Saimir Osmani (Civil and Risk Engineer, RINA), the available sample datasets and machine-learning models developed within the MEDiate project were reviewed. Following a technical assessment, it was concluded that the historical European flood datasets and associated modelling approaches were not sufficiently representative of Essex's local weather, hydrological and infrastructure characteristics to support scientifically robust machine-learning model development. Consequently, both parties agreed to redirect the technical focus of the secondment.

Stage 3 – AI-assisted hazard intelligence and resilience planning

Following the change in direction, the remainder of the secondment was undertaken independently by Ser-Huang Poon. The work explored the feasibility of using Large Language Models (LLMs), Google Maps and AI-assisted information retrieval to identify, collect and organise publicly available information relevant to flood resilience in Essex. Prototype hazard-event templates, representative inventories of critical infrastructure and vulnerable communities, and example flood and drought case studies were developed to demonstrate how AI could support the creation of future datasets for resilience modelling and emergency planning.

Throughout the secondment

Weekly meetings were held with the RINA team to discuss progress, present interim findings and obtain technical feedback. These discussions helped shape the evolving research direction and ensured that the work remained aligned with the practical needs of resilience practitioners while benefiting from RINA's expertise in infrastructure resilience and disaster risk management.

Final phase

Concepts developed during the secondment—including AI-assisted hazard intelligence, human-centred resilience planning, federated data architectures and agent-based modelling of vulnerable communities—were subsequently incorporated into the INTRA-CARE Horizon Europe proposal. Ser-Huang Poon 

Outcomes/outputs

1. INTRA-CARE (Interscalable Resilience Assessment for a Secured Healthcare Infrastructure System), led by RINA, was submitted to the Horizon Europe Innovation Action programme to develop an AI-enabled resilience framework for European healthcare infrastructure.

   The proposal involved a 14-partner European consortium with a total eligible project budget of €5.62 million and a requested EU contribution of approximately €5.0 million. The University of Manchester was allocated a budget of €234,085 (approximately 4.7% of the total project budget). Building on the SPRITE+ secondment, the University contributed expertise in trustworthy AI, AI-assisted hazard intelligence, federated data architectures, vulnerable community modelling and agent-based resilience planning. Although the proposal exceeded the funding threshold, it was not selected because reviewers considered the project overly ambitious and insufficiently detailed in demonstrating toolkit integration, methodological validation and the achievability of its proposed outcomes.

2. Poon, S.-H. and Wang, C. (2026) An AI framework for measuring organisational cyber resilience from corporate disclosures: Towards trustworthy digital infrastructure. Working paper submitted to the International Conference on Trustworthy Digital Infrastructure 2026, The Alan Turing Institute, London, UK, 15–16 September 2026.

   This paper directly builds on experience gained during the SPRITE+ secondment. The secondment demonstrated that AI's greatest value lies not simply in prediction, but in collecting, structuring and interpreting complex, heterogeneous information from multiple knowledge sources. The Essex resilience work explored the use of LLMs, AI-assisted information retrieval and structured knowledge engineering to organise hazard intelligence for resilience planning. These concepts evolved into the paper's AI architecture, where corporate disclosures are combined with external knowledge bases and regulatory guidance through RAG and an explicit resilience ontology. The emphasis on explainability, traceability, evidence grounding and human-centred decision support reflects key lessons learned during the SPRITE+ collaboration with RINA.

3. Wang, C., Poon, S.-H., Wei, S. and Tsado, Y. (2026) Capturing cyber risk in the UK equity market.

   Working paper to be presented at the 3rd Financial Fraud, Misconduct and Market Manipulation Conference, Lancaster University Management School, Lancaster, UK, September 2026. This paper examines whether cyber risk is priced in the UK equity market and how its pricing varies according to firms' digital dependence, supply-chain vulnerability, governance quality and operational resilience. The underlying methodology builds directly on ideas developed during the SPRITE+ secondment, extending AI-enabled resilience assessment from critical infrastructure to financial resilience and cyber risk. The secondment demonstrated how AI, Large Language Models (LLMs) and knowledge engineering can efficiently collect, organise and interpret diverse information from multiple public sources to support resilience analysis. Building on this foundation, the paper develops an AI pipeline that extracts cyber-related evidence from annual reports, enriches it with regulatory guidance (FCA, FRC and NCSC) and cybersecurity knowledge (MITRE ATT&CK) through Retrieval-Augmented Generation (RAG), and constructs transparent, explainable measures of cyber risk for asset pricing. The resulting framework demonstrates how trustworthy AI and evidence-based knowledge integration can support more robust resilience assessment and informed decision-making.

4. Keraminiyage, K., Ingirige, B., Newbery, S., Poon, S.-H. and Tsado, Y. (2026) Quantifying the power–communications deadlock: A mixed-methods calibrated simulation of cascading failure in the UK energy–communications interdependency. Working paper for submission to the International Journal of Disaster Risk Reduction.

   This working paper develops a calibrated simulation framework to investigate cascading failures between the UK electricity and communications sectors. Using a mixed-methods approach that combines a systematic literature review, expert Delphi elicitation and a 21-node interdependency simulation, the study examines how cyber and physical disruptions propagate across interconnected critical infrastructure. The results identify the power–communications deadlock as the dominant systemic failure mode, demonstrate the highly non-linear nature of cascade dynamics, and show that interventions such as SCADA segmentation, extended communications battery backup and improved governance can significantly enhance system resilience.

   Although this paper is only indirectly related to the SPRITE+ secondment, it builds on one of its central themes: understanding how failures propagate across interconnected infrastructure systems. The secondment reinforced the importance of analysing multiple infrastructure dependencies, cascading effects and resilience planning from a systems perspective. These concepts are reflected in the paper's simulation framework, which integrates technical interdependencies, governance and recovery processes to evaluate resilience interventions. While the simulation methodology itself was developed independently, the SPRITE+ collaboration helped shape the broader research direction towards multi-sector resilience assessment and evidence-based decision support.

Impact

Although exploratory in nature, the secondment has generated impacts that extend beyond its original objectives and established a foundation for longer-term collaboration between academia and industry in AI-enabled resilience assessment.

Organisational impact

The collaboration enabled the University of Manchester and RINA to evaluate where AI can add value to critical infrastructure resilience. An important outcome was the recognition that AI's greatest immediate contribution lies not in replacing existing resilience models, but in improvingthe collection, organisation and interpretation of diverse hazard information needed to support credible modelling and decision-making. The project demonstrated the feasibility of using Large Language Models (LLMs), Google Maps and AI-assisted information retrieval to rapidly assemble structured hazard intelligence from multiple public sources. This approach has the potential to reduce the time and effort required to prepare evidence for resilience assessments while improving transparency and traceability.

Policy and sector impact

The secondment reinforced the importance of understanding infrastructure interdependencies, cascading failures and human-centred resilience within the UK critical infrastructure sector. The Essex case studies illustrated how AI can support the systematic development of locally relevant hazard scenarios, critical infrastructure inventories and evidence bases for emergency planning. The work also highlighted the need to combine AI with trusted public information and domain expertise, supporting the wider adoption of trustworthy AI in resilience planning rather than relying solely on predictive machine-learning models.

Research and innovation impact

The collaboration created significant follow-on research opportunities. Concepts originally explored during the secondment—including AI-assisted hazard intelligence, federated resilience data architectures, vulnerability assessment and agent-based modelling of cascading failures—were further developed within the Horizon Europe INTRA-CARE proposal led by RINA, with the University of Manchester joining the consortium as a collaborative partner. Although the proposal was ultimately unsuccessful, it established new international collaborations and provided a platform for future funding opportunities. The secondment also informed several ongoing research projects applying AI, retrieval-augmented generation (RAG) and knowledge engineering to cyber resilience, disaster risk reduction and trustworthy digital infrastructure.

Social impact

The project increasingly adopted a human-centred perspective by considering how infrastructure failures affect vulnerable communities. The AI-assisted methods explored during the secondment have the potential to support emergency planners in identifying vulnerable populations, understanding cascading impacts across essential services, and improving preparedness for future flood and drought events.

Overall, the secondment demonstrated that AI could make an important contribution to resilience engineering by helping organisations transform fragmented information into structured, evidence-based knowledge. While the work was a feasibility study rather than a full operational deployment, it established practical AI workflows, strengthened strategic collaboration

Future work

The SPRITE+ secondment established a strong foundation for future research on the use of trustworthy AI to strengthen resilience and cyber security. Building on the feasibility work completed during the secondment, the project team has prepared a proposal for the CRANE Phase One Pilot Projects Call, which investigates how frontier Large Language Models (LLMs) can assess trust, provenance and coordinated influence operations within digital information ecosystems.

The proposed research extends the AI-assisted knowledge engineering developed during the secondment from physical infrastructure resilience to cyber resilience. Instead of analysing hazards such as flooding, it explores whether AI can identify coordinated misinformation campaigns and AI-generated content before harmful narratives become established. By orchestrating large numbers of LLM analyses, the project will develop interpretable Authenticity, Coordination and Risk scores as early-warning indicators of suspicious online activity.

The project will strengthen collaboration between the University of Manchester, the University of Salford and CRANE partners, creating opportunities for researchers, industry and government organisations to contribute to the development and evaluation of trustworthy AI methods. Open-source software, validated methodologies and, where appropriate, annotated datasets will be shared to support wider adoption and future research.

More broadly, the AI-enabled approaches explored during the secondment—including knowledge extraction, retrieval-augmented analysis and evidence-based resilience assessment—will continue to underpin research on critical infrastructure, cyber resilience and trustworthy digital infrastructure, providing a platform for future UKRI, EPSRC and Horizon Europe collaborations.