Police in Auckland have admitted that a staff member used artificial intelligence to alter a photograph before it went on an official police Facebook page, and that the edit changed what the object in the picture appeared to be.
The photograph was posted on Wednesday to the North Shore, Rodney and West Auckland Police page. It showed an officer and three workers at a Takapuna music store holding a bass guitar that had been recovered more than three years after it was stolen. The post said the instrument was a Fender electric worth $4,299.
People who looked at the picture did not believe it. The guitar in the image carried a Squier logo, and a Squier J Bass sells for a small fraction of that price. Squier is Fender’s budget brand, and the gap between the two is the difference between a first instrument and a professional one. Commenters said so bluntly. “$4k for a Squier Bass… I’ve got a bridge I can sell you, real cheap,” one wrote. “That’s a Squire only worth $600,” wrote another, according to RNZ, which first reported the story on Thursday evening.
The manager of the Takapuna store told RNZ the stolen instrument was in fact a Fender Jazz American Bass, and that police had used AI on the photo. When he asked what had happened, he was told the bass had looked “too blurry” in the original shot, so AI had been used to try to make it look better. The tool did not simply sharpen the image. It put a Squier logo on the headstock of an instrument that never carried one.
Police deleted the post after RNZ sent questions about it. In a statement, police said the photograph was “genuine” but confirmed AI had been used on it. “Unfortunately, a staff member used AI to enhance the focus of the photo,” the statement said. “This goes against police processes and should not have happened. The post has been removed.” RNZ has asked police for the original, unaltered image and for detail on how much of the picture was changed.
The recovery itself was real police work. Officers were alerted in May to a guitar listed on Trade Me with a serial number matching the stolen one, and traced it back to the person who had taken it. The instrument was returned to the store in reasonable condition, though the manager said it would now be resold for considerably less than it was worth before.
What makes this more than a small embarrassment is that police have a written rule covering exactly this situation. Under the police acceptable use of generative AI policy, staff using a generative AI tool are meant to label the output as AI generated and name the tool used, unless there is a clear operational reason to keep that quiet. “The label can vary to reflect the level of involvement of the genAI tool and the human,” the policy says. The policy says being transparent about generative AI use is important to “help preserve the trust between the person writing/developing content and the audience”.
That is the awkward part. A police Facebook page is not a marketing channel. It is one of the places where the public is asked to accept, without independent verification, that a photograph shows what the caption says it shows. Police post images of stolen property, of vehicles, of people they want to identify, and of scenes they want the public to recognise. The whole value of those posts rests on the assumption that nothing in the frame has been invented. An image-sharpening tool that quietly adds a brand name to a guitar is, in principle, the same tool that could add or remove a detail on a number plate, a jacket or a face.
Police have themselves been warning the public about precisely this problem. In April, after a Facebook page shared an AI-generated image purporting to show body bags at the scene of a triple homicide in Hastings, police and Netsafe both spoke out. Police executive director of media and communications Cas Carter told the New Zealand Herald that police were deeply concerned about the ways technology was being used to depict crime scenes, and that one of the main problems was telling AI-generated images apart from real ones. She said there were limited legislative controls on such content. Netsafe chief online safety officer Sean Lyons said in the same story that “Being upfront about its origin helps prevent harm.”
It is not the first time police AI use has come under scrutiny this year. In June, RNZ reported that police had restarted their use of the Whisper transcription tool in September 2025 after shutting an earlier trial down in March 2025 over misuse, with police acknowledging that switching the tool off had increased the risk of staff reaching for unapproved alternatives instead. That is close to what appears to have happened here. Nobody set out to mislead anyone about a bass guitar. A staff member wanted a clearer picture, reached for a tool that was easy to get at, and did not understand that modern image tools do not sharpen images so much as redraw them.
That is the technical point worth holding on to. The consumer AI features now built into phones, photo apps and messaging software are not doing what the old sharpen slider did. They are generating new pixels based on what the model thinks should be there. Given a blurry headstock on a bass guitar, a model trained on millions of photographs of bass guitars will fill in a plausible logo. Plausible is not the same as true, and there is no warning when the two part company.
New Zealand schools have already started pulling back for related reasons, with two Auckland schools this week stopping the posting of student photographs to Facebook over concerns about how those images could be reused. The police case points at a different risk from the same technology, which is not what other people do with an official image but what an official image has quietly had done to it before anyone else sees it.
Police have not said which tool was used, whether the staff member had completed the training required before accessing approved AI tools, or whether any other posts have been altered the same way. Those answers matter more than the guitar does. An organisation that asks the public to believe its photographs needs to be able to say, without hesitation, that its photographs are photographs.
Should police be required to label every image they post that has been touched by AI, or should they simply not use these tools at all on operational photographs? Have your say in the comments below.
This article was written by AI, briefed to report the facts, hopefully without some of the bias people bring to the job π
Leave a Reply