Emerging and disruptive technologies can significantly influence national cyber-defence capabilities. Developments in artificial intelligence, quantum technologies, advanced communication networks, confidential computing, autonomous systems, and privacy-enhancing technologies may create new defensive opportunities, strategic dependencies, and security challenges.

Relevant information is distributed across scientific publications, patents, standards, technical reports, funding announcements, and industry news. The growing volume and diversity of these sources make manual monitoring increasingly difficult. Conventional technology-monitoring approaches—such as keyword searches, publication counts, and bibliometric analysis—can detect established trends but may struggle to identify weak signals, semantic relationships, and early convergence between previously separate technological fields.

Large language models (LLMs), semantic embeddings, retrieval-augmented generation (RAG), and knowledge graphs offer new possibilities for analysing large and heterogeneous document collections. These methods could help identify emerging technologies, connect related developments, explain why a trend may be strategically relevant, and trace conclusions back to supporting evidence. However, it remains unclear whether LLM-based approaches can detect meaningful developments earlier and more reliably than conventional methods. They may also generate unsupported connections, amplify technological hype, or confuse frequent discussion with genuine technological progress.

Objective:

This thesis will investigate the capabilities and limitations of LLMs for early-warning and technology-convergence monitoring in the context of national cyber defence.

The objective is to design and evaluate an LLM-assisted monitoring framework that detects early signals of emerging technologies and identifies potentially significant convergence between technology domains relevant to national cyber defence.

The emphasis will be on evaluating whether LLM-based methods provide measurable advantages over conventional monitoring techniques—not simply on developing a technology-monitoring dashboard.

Research questions:

• Can LLMs detect emerging technological developments earlier or more accurately than keyword-frequency and bibliometric approaches?
• Which information sources provide the earliest and most reliable indicators of technological progress?
• Can LLMs distinguish genuine technological developments from speculation, marketing claims, and short-term hype?
• Can LLM-based knowledge graphs identify meaningful convergence between previously separate technology domains?
• How can genuine technology convergence be distinguished from coincidental co-occurrence?
• How stable are the identified signals across different models, prompts, document collections, and time periods?
• How accurately can the system link detected signals and relationships to supporting evidence?
• How should uncertainty and confidence be represented in an early-warning system?

Possible case-study domains:

The student should select one or two focused cases, such as:

• Artificial intelligence
• Post-quantum cryptography
• Quantum technologies
• Autonomous agents
• Space Cybersecurity

Expected contributions:

The thesis is expected to produce:

• A structured methodology for LLM-assisted monitoring of emerging and converging technologies.
• A time-stamped dataset covering one or two national cyber-defence technology domains.
• A comparison between LLM-based and conventional monitoring approaches.
• An evaluation framework for measuring detection quality, timeliness, evidence grounding, and stability.
• A prototype technology radar showing emerging signals, convergence patterns, supporting evidence, and confidence levels.
• Practical recommendations on where LLMs are—and are not—reliable for strategic technology monitoring.