Whole Of Case Visibility
Whole of Case Visibility is a principle requiring comprehensive transparency and forensic-level auditability in AI systems deployed in high-stakes decision-making contexts, particularly within criminal justice. The principle emerged from the Farah Jama case, in which an Australian man was wrongfully arrested based partly on a face recognition match that lacked adequate human scrutiny and documentation. The case exposed critical gaps in the visibility of how AI systems contributed to decisions affecting individual liberty, highlighting the need for complete documentation and traceability throughout investigative and adjudicative processes.
Core Requirements
The principle demands that every stage of AI involvement in consequential decisions be documented, explainable, and subject to independent review. This includes the data inputs, algorithm outputs, confidence scores, and the reasoning process by which human decision-makers incorporated or disregarded AI recommendations. Unlike general transparency principles, Whole of Case Visibility specifically emphasizes the capacity for post-hoc forensic examination—the ability to reconstruct exactly what information and reasoning led to a particular outcome, even months or years after the decision was made.
Application in Criminal Justice
The relevance of Whole of Case Visibility extends beyond the original case to any criminal justice context where algorithmic tools influence arrest, bail, sentencing, or parole decisions. The principle responds to the structural problem that AI systems can influence outcomes without leaving an adequate evidentiary trail, making it difficult to identify errors, bias, or procedural failures during appeals or reviews. Implementation requires both technical measures—such as comprehensive logging and model documentation—and institutional practices that ensure human accountability remains central to high-stakes decisions.
Source Notes
- 2026-04-14: I Looked At Amazon After They Fired 16,000 Engineers. Their AI Broke Everything.