Unified Guidance Is All You Need: Re-Engineering New Zealand Government Guidance for Trustworthy AI System Delivery

The main challenge is in the implementation gap. The guidance giving effect to these laws and offering best practice is currently fragmented, making it difficult for agencies to translate high-level commitments into trustworthy operational practice. To improve the delivery of trustworthy AI systems, this guidance should be consolidated.

Building on an analysis of existing techniques and applications, I devise a taxonomy that categorises algorithms and AI systems in sufficient detail to balance technical specificity and operational practicality. Techniques are grouped according to the evaluation method they employ. System use cases are grouped based on their broad risk profile: those making individual conclusive decisions, those generating the evidence base for broad policy decisions, and those automating core job functions.</p><p>This taxonomy underpins a proposed New Zealand Artificial Intelligence Manual (NZAIM).

Designed to reflect the structure of the existing New Zealand Information Security Manual (NZISM), the NZAIM should provide a unified catalogue of best practice to promote trustworthy delivery of Government AI. Just as the NZISM defines the policies and procedures essential for a robust, enduring information security culture, the NZAIM should promote trustworthy AI system delivery not as a mere checklist but as a continuous commitment to safety, effectiveness, and accountability that builds lasting trust in Government AI.</p>