No law, no mandate. New Zealand asked its agencies to sign a voluntary charter on how they use algorithms. Five years on, 29 have signed — and the question is whether voluntary commitments actually change practice.

The dilemma

By 2019, New Zealand's government was using algorithms across welfare, immigration and policing faster than its laws could keep up. Writing a new algorithm law risked freezing the technology. Doing nothing risked a different problem: algorithms making consequential decisions about citizens without the public being able to see how, or hold anyone accountable when things went wrong.

The regulatory gap

A way to make government use of algorithms transparent and accountable without writing new law. Existing legislation, including the Privacy Act and the Official Information Act, covered parts of this but left the questions citizens cared about unanswered: which decisions involve algorithms, who is responsible for them, and how can outcomes be challenged?

The approach

In July 2020, Stats NZ launched the Algorithm Charter for Aotearoa through its Government Chief Data Steward. The Charter asks signatory agencies to commit to five things: transparency, partnership with Māori, data quality, peer review, and human oversight. The thinking was simple. A visible public commitment, backed by peer support and an independent review after one year, would shift agency behaviour faster than passing a law.

Results achieved

Twenty agencies signed at launch; by July 2023 there were 29 signatories, including Police and the Ministry of Business, Innovation and Employment (MBIE). The independent Year One Review by Taylor Fry (2021) found near-universal support among agencies and experts, with documented practice shifts toward greater transparency.

The Charter has limits, its designers acknowledge. It is voluntary, public-sector only, and not centrally measured — there are no consolidated public metrics on algorithmic error rates, bias reduction or service outcomes. Stats NZ released an Algorithm Impact Assessment toolkit in December 2023 to give signatories more practical scaffolding.

The transferable lesson

Voluntary frameworks need three things to work. A credible agency to convene the participants. A community small enough that people actually know each other. And a built-in review point that creates accountability without formal enforcement. Without all three, voluntary frameworks tend to fall flat, which is why most attempts to copy this model elsewhere have not produced the same change in behaviour.