New Zealand Police Misused AI Translation Tools, Documents Show

New Zealand police misused AI translation tools, including uploading 350,000 lines of Chinese audio to Google Translate, documents released to RNZ show.

maisiekooc
Maisie Morrison

AgentLocker Editor

AI News
New Zealand Police Misused AI Translation Tools, Documents Show

New Zealand police investigators misused artificial intelligence translation tools several times in risky ways, according to official documents released to RNZ.

In one case, an investigator uploaded 350,000 lines of Chinese audio data to Google Translate. Police consider that service an unsecured platform.

The documents include hundreds of pages of emails and reports released under the Official Information Act. Police declined an interview request from RNZ.

How Whisper AI Entered Police Work

The use of AI began during Operation Parade, the 2024 Auckland investigation into the killing of Shulai Wang. Her body was found in a bag in Gulf Harbour in July 2024.

Officers had intercepted phone calls and collected 250 hours of audio in Mandarin Chinese, spread across more than 2,500 files. An email said the investigation had interpreters, but their time was limited.

Investigators also warned that several targets had become a higher flight risk. They turned to Whisper, an AI system, to speed up the work.

Whisper finished in 10 hours a job that would have taken a person about 750 hours and cost around $50,000. All evidence still had to be checked later by qualified human translators.

After this, automated translation spread to other police teams. Staff who did transcriptions became worried about what it meant for their jobs.

Rules Broken During the Trial

Police set up a six-month formal test of Whisper with controls. Officers were not allowed to use it on English audio or on Māori or Pacific languages.

An evaluation found officers broke the no-English rule many times. In one case, Waikato investigators used AI to create a document for disclosure in legal proceedings.

Detective Inspector Stu Mills, who manages Intercept and Technology Operations, said the breaches posed serious security and privacy risks. He also called them “unfortunate but not unexpected” because of the pressure on investigators.

Some investigators took evidence home to process on personal devices. A police report said this risked “exposing confidential information.”

Whisper was dropped in March 2025, except for a few one-off cases that needed three levels of approval. Several of those involved transcribing English from hundreds of prison phone calls.

Investigators kept using AI anyway. In one case, they used Cellebrite, a phone-cracking tool, to pull WeChat threads from phones before putting the Chinese audio on Google Translate.

They wrote that without sorting the data quickly, the investigation could run “for years damaging the police’s reputation.” The case was not named but was likely a 2025 financial investigation.

Staff also grew frustrated with changing rules. One High Tech Crime Group manager wrote, “We are exhausted.”

Mills said the lack of a secure, approved tool had pushed staff toward “these risky workarounds.” He called for a trusted in-house option.

Police reintroduced Whisper about a year ago. By October 2025, it was being built into Cellebrite, which police already used.

The documents suggest Whisper now appears widely available to officers. It remains unclear how its use is audited or how the rules are enforced.

From our research desk
AI Jobs Automation Index
Which jobs are AI tools targeting most? We mapped 3,400+ AI tools to real job functions — with BLS employment & salary data.
Explore the index
maisiekooc

Written by

Maisie is a news writer at Agent Locker, covering the latest developments in artificial intelligence, emerging technology and the companies shaping the future.

Discover AI Agents