New Zealand's Inland Revenue Told Its Minister That an AI Tool's Call Summaries, Once Checked for Completeness and Correctness, Can Be Copied to the Customer Record

In March 2026 New Zealand's Minister of Revenue released four reports that Inland Revenue had sent the minister during 2025. Each is an update on Inland Revenue's use of artificial intelligence, and they are dated February 26, May 9, August 27 and December 5, 2025. In one of them, a paragraph was withheld under the Official Information Act 1982 and released in the form of a summary. Parts of the reports follow, not the whole.

An AI Tool for Call Summaries

The February report describes a proof of concept of a tool that creates summaries of calls between a customer and Inland Revenue. The tool uses the call transcript and generates a summary of the key points "which, once checked for completeness and correctness, can be copied to the customer record." And the report adds, "Note the AI generated summary is not retained."

The proof of concept, the report says, demonstrated an average 17% decrease in after-call work time and "no degradation in quality of notes." The plan it set out was to roll the tool out to one team and to continue pilots with other business areas "to ensure there is value across other customer cohorts before further rollout."

Assurance

An appendix to the same February report lists reflections and learnings. Among them are "Having assurance models and measurement frameworks" and, under "Beware of AI Washing," the line "Assuming you have AI in product/platform without investigating it fully, could lead to you applying the wrong assurance yardsticks."

The May report says Inland Revenue officials met with the Government Chief Digital Officer that quarter to discuss a proposed Public Service AI Assurance Model. "Inland Revenue Officials agree with the need to have a public service AI assurance model and provided feedback on changes to balance risk while enabling government organisations to rapidly innovate and adopt AI."

In August, Inland Revenue reported that it had consulted on an OECD project, "Enhancing the trustworthy use of AI in Tax Administrations," whose draft report presented "a draft assurance checklist specific to AI use cases known across tax administration sectors." It shared feedback confirming its support for the material that accompanies the checklist. "However, the draft assurance checklist is overly detailed, which may hinder swift progress." It said it had proposed specific changes "to allow tax administrations to adjust their pace based on risk."

Risks and a Review of the Law

In May, the report said, "Inland Revenue's key risk remains that of data and information leakage." It also said an initiative had commenced "to test and monitor public GenAI engines" for "accuracy and consistency of answers to questions the public may ask about complying with tax obligations or obtaining social entitlements."

The same report says the Chief Tax Counsel has initiated a review to determine whether the Tax Administration Act 1994 "is currently fit for purpose for digital advancements including AI," and that the review will consider Inland Revenue's interpretation of the legislation and whether legislative changes may better enable it to use AI capabilities.