Generative artificial intelligence (AI) tools such as large language models are now used across the research cycle, from literature searching and coding qualitative data to drafting manuscripts. Their use promises efficiency but also raises problems of fabrication, bias, confidentiality, authorship and accountability. This article examines how ethical principles are, and are not, upheld when AI is used in formal research investigation, meaning the systematic, methodologically governed inquiry that produces knowledge claims for scholarly scrutiny. A qualitative multiple-case study design was used. Four publicly documented cases of AI use in research and scholarly publishing were analysed through document analysis and reflexive thematic analysis, and were compared against international ethics and integrity guidance. An illustrative composite case of a university research team is added to show how the identified themes operate in everyday practice. Five themes were developed: (1) verification as the central ethical duty; (2) transparency and disclosure gaps; (3) accountability that cannot be delegated; (4) confidentiality and data protection; and (5) bias, equity and the erosion of researcher judgement. Failures arose less from the technology itself than from weak human oversight and unclear institutional rules. Ethical use of AI in research depends on a human-in-the-loop model supported by disclosure, verification protocols, data governance and training. A practical framework (the V-T-A-C-J model) is proposed.
Keywords: Generative artificial intelligence (AI, Ethics of AI, Framework of AI in research