Since the release of ChatGPT in late 2022, adoption has accelerated while concerns persist regarding security and resilience, data protection, and harmful content outputs. Existing red-teaming and safety testing tools provide partial coverage but do not support an integrated, comparable assessment across services.

Using a Design Science Research approach and scoped to the technically measurable dimensions of cybersecurity and resilience, data protection, and content safety, this thesis developed (1) a Digital Trust framework specifying dimensions, criteria, evidence types, and scoring rules, and (2) a proof-of-concept benchmark that operationalises the framework by synthesising existing tools. The work is informed by a systematic literature review across DBLP and Scopus; database searches returned 4,068 records; following title and abstract screening, a final corpus of 16 studies informed the framework design.

The benchmark was applied to three LLM-based textual AI services, producing differentiated comparative Digital Trust scores and demonstrating the feasibility of integrated, practitioner-oriented trust assessment.