Police say they're assessing how to use emerging technology after report of 'risky' AI use

Police say they are continually assessing how they can use emerging technology to support investigations in a lawful and secure way consistent with their policies.

Official documents show police broke their own safeguards multiple times using generative AI in investigations in 2024 and 2025.

At the same time, the superfast AI translation of phone and online communications sped up criminal investigations including into the Auckland cult killing of Shulai Wang.

RNZ has revealed police used both a system they had vetted for use called Whisper AI - but broke their own rules for it - as well as using unsecured systems such as Google Translate in what their own leaders called "scary" and "risky workarounds".

With Whisper, frontline investigators also used it to translate te reo Māori conversations, which broke police rules for the test, the documents showed.

Detective Inspector Stuart Mills of the National Criminal Investigations Group told RNZ there was strong operational interest in using tech.

"Investigators often deal with fast-moving and complex investigations involving significant volumes of foreign-language material," he said in a statement on Monday.

The documents released under the OIA showed Mills argued last year that officers should be allowed to keep using Whisper AI even though they broke the rules repeatedly in a six-month test of it, using it to transcribe English audio when it was only for use for translating.

"Introducing additional controls to restrict the use of the service would likely hinder its effectiveness," Mills said in a report in March 2025.

Automated tools for transcription and translation were not just helpful, but "essential".

Barrister and privacy law expert Kathryn Dalziel told Checkpoint on Monday it was "absolutely unacceptable" that police took data home and loaded it up into open source systems.

Though Whisper was largely suspended in March because of the misuse, police started it up again last October.

Mills said on Monday they had to balance the opportunity to identify critical information more quickly with important privacy and security considerations.

He added police had "developed approved solutions" but did not name Whisper or any other AI tool.

The force's latest Technology Capabilities List of October 2025 does not refer to "translation".

Mills said police took a considered approach to adopting new tech, deciding according to operational need, risk, legal requirements and policy considerations.

They had robust testing, assurance and approval processes to support "effective policing while maintaining public trust and appropriate privacy protections".

Police began using Whisper AI in 2024 without having detailed auditing controls over it though they agreed the logs of how it was used should be expanded to include more information for auditing and analytics, according to the OIA reports and emails.

However, by July 2025, in a discussion over how Whisper was being incorporated into a phone-harvesting tool they already used called Cellebrite, the police's security and privacy reference group wrote: "There are currently no SOPs [standard operating procedures], audit controls, or training materials governing its use, and its evidential limitations must be clearly communicated to staff."

As for making any reviews of it available to the public, the group's report said: "No assessments or reviews are being considered for these new functionalities."