2026-10-13T18:17:36Z https://api.figshare.com/v2/oai

oai:figshare.com:article/34015221 2026-09-29T15:29:00Z category_28864 portal_543 item_type_8 month_year_09_2026

Advancing a Comprehensive Understanding of AI-Related Risk: From Individual Studies to Collective Insights Shasha Yu (12883613) Artificial intelligence not elsewhere classified Artificial Intelligence AI Risk AI Safety Responsible AI <p dir="ltr">Artificial intelligence introduces complex and multifaceted risks that are often examined within particular technical, institutional, governance, or human contexts. This fragmentation creates challenges for understanding how different forms of AI-related risk are represented across contexts and analytical levels. This thesis addresses this problem through a programme of seven published studies and a retrospective cross-portfolio synthesis of their findings.</p><p dir="ltr">The studies examine AI-related risk across national security, policing, AI privacy-protection research, human–AI trust, smart home environments, the operational behaviour of LLM-based agents, and AI risk understanding and judgement among university students. Three overarching research questions examine how AI-related risks manifest across institutional and social contexts; structural characteristics and gaps in AI privacy-protection research; and patterns across the portfolio and the further insights they provide into AI-related risk.</p><p dir="ltr">The portfolio provides different forms of evidence across these questions. Papers 1, 2, 4 and 5 provide institutional, conceptual and framework-based perspectives on AI-related risk. Paper 3 identifies substantial growth in AI privacy-protection research alongside geographical concentration, technically dominant research trajectories, and comparatively underdeveloped human-centred, ethical and legal dimensions. Paper 6 identifies divergence between policy-aligned language and executed action in LLM-based agents, while Paper 7 identifies a gap between stated risk awareness and applied risk judgement. Comparison across the portfolio further identifies recurring patterns concerning surface indicators, blind spots in risk assessment, the multidimensional character of AI-related risk, and the different analytical levels represented across the studies.</p><p dir="ltr">The cross-portfolio synthesis organises these findings through two complementary perspectives: a Meta–Macro–Meso–Micro analytical structure and a comprehensive conceptualisation comprising five interconnected risk domains. These perspectives are organisational and descriptive syntheses grounded in the domains examined in the portfolio. They provide an analytical reference for considering AI-related risk across different levels, contexts, and dimensions, while their applicability to additional domains requires further validation and may require extension or revision. Together, the portfolio provides a comprehensive but bounded synthesis of AI-related risk across the domains examined. </p><p><br></p> 2026-09-29T15:29:00Z Text Thesis 10.25401/cardiffmet.34015221.v1 https://figshare.com/articles/thesis/Advancing_a_Comprehensive_Understanding_of_AI-Related_Risk_From_Individual_Studies_to_Collective_Insights/34015221 CC BY-NC-ND 4.0