Most AI ethics frameworks hold up on paper. The gap is accountability.

Most AI ethics frameworks in healthcare hold up well on paper. The gap shows up in one place: accountability.

That was the finding of my Master of Research (Macquarie University, Australian Institute of Health Innovation). The case study applied the NSW AI Assurance Framework to a real system — an AI tool used in wound care — using the SEIPS work-systems model to look at how the framework actually operated around clinicians, IT and governance, not just what it said on paper.

Where governance already works — and where it doesn't

Existing health governance already covers most AI ethics principles well: community benefit, privacy, security, transparency. Fairness and accountability are the exceptions. Accountability especially — in a hospital, no single person owns it. It sits across clinicians, vendors, IT and governance committees: a mosaic of accountabilities, not a clean chain of responsibility.

What helps, and what gets in the way

Facilitators were specialist knowledge, multidisciplinary collaboration and strong governance processes. Barriers were time, resourcing, and navigating that accountability across stakeholders.

The thesis is open access: doi.org/10.25949/28585217