New Zealand's Parliamentary Counsel Office (PCO), the agency that drafts all government legislation, ran a six-month research-and-development programme in 2024–2025, partnering with five technology companies (including Catalyst IT) to explore generative-AI applications in legislative drafting before making any decision to adopt the technology.
The team tested at least four distinct use cases: drafting clause-by-clause explanatory notes for new bills; generating prospective (draft-updated) consolidations of amended Acts; supporting plain-language legislative writing; and prototyping a chatbot for the New Zealand Legislation website. For the explanatory-notes and consolidation work, PCO fine-tuned a small, 8-billion-parameter open language model on around 200 examples of New Zealand legislative text, rather than relying on a large third-party foundation model.
To keep sensitive legislative data within New Zealand jurisdiction, the proof-of-concept models were deployed on Catalyst Cloud, New Zealand's own sovereign cloud infrastructure. PCO published candid, per-experiment results pages (for example, "How successful is AI at drafting an explanatory note?") rather than only promotional summaries, and explicitly framed the work as exploratory — assessing benefits, risks and limitations before any production rollout. The pilot follows New Zealand's earlier, non-AI "Better Rules"/"Rules as Code" programme (2018 onward), which OECD case studies credit with international influence but which practitioners say stalled for lack of senior sponsorship — a contrast that makes PCO's current, more cautious AI trial notable.
Read the full analysis: https://www.pco.govt.nz/about-us/legislative-data-and-technology
Where this practice's information was retrieved from, and when.