Research / Ethics
Ethics
I developed the confidence-scoring methodology for systematic ethics-based audit of large language models. Subsequent work extends the audit framework to negation sensitivity and value alignment. NIST audit and reporting work continues in this register.
The confidence-scoring audit methodology
- “Informed AI Regulation: Comparing the Ethical Frameworks of Leading LLM Chatbots” (Chun and Elkins, arXiv:2402.01651, 2024) — introduces the confidence score, a scalar measure of normative certainty for evaluating model hesitation versus firm commitment in moral reasoning. To the authors' knowledge, the first systematic ethics-based audit of LLMs.
The methodology has been directly adopted in subsequent NLP work: Liu, Liu & Yu (COLING 2025) use the confidence score across 1,613 social decision-making scenarios; Jain, Calacci & Wilson (AIES 2024) cite the audit framework; the Ungless et al. LLM Ethics Whitepaper (2024) treats it as a representative ethics-based audit; Snoswell, Kilov & Lazar (AAAI 2026 AI Alignment Track) include it as one of 69 foundational works in their “Beyond Verdicts” survey.
Negation, prohibition, and value alignment
- “When Prohibitions Become Permissions: Auditing Negation Sensitivity in Language Models” (Elkins and Chun, arXiv:2601.21433, 2026) — examines how LLMs misinterpret negated moral statements, with “should not” frequently producing permissive rather than prohibitive responses. Cross-listed with Language.
Institutional ethics work
- IBM–Notre Dame Tech Ethics Lab grant recipient (also Recognition)
- NIST CAISI ethics-task-force contributor (cross-listed with Governance)
- Public NIST submissions on AI ethics standards