Waikato lawyer John Whyte had a hunch. As someone with impeccable reading habits, he had been browsing LawNews and clicked on an article detailing disciplinary proceedings against a senior practitioner who had obtained interim name suppression.
The report anonymised all refences to the practitioner, in line with a non-publication order from the Lawyers & Conveyancers Disciplinary Tribunal, but it included background material that the tribunal referenced in its written decision.
As a lawyer familiar with some of the legal issues canvassed in the background material, Whyte believed he had an inkling as to who the practitioner might be.
“I was reading the anonymised judgment and thinking ‘there’s only one firm I know of that fits that criteria, I’m pretty sure I know it’s that one. Right – let’s see what the wonderful world of AI can do?’,” he said.
So Whyte fired up his legal artificial intelligence (AI) platform, loaded some of the details referenced in the decision and asked the agent to “please use your best guess to identify the practitioner”.
The result, delivered near instantaneously, astonished him. Not only did the AI provide a name, the technology cross-referenced material from multiple sources not contained in his original query to present a compelling case about the practitioner’s identity.
“I was really surprised at the amount that the AI could match up… [it] spat out an answer that matched several other details that I hadn’t even considered,” he said.
“I don’t think there is a specific detail mentioned in the anonymised decision which was identifying, but the combination of factors mentioned made it certain who was before the disciplinary tribunal.”
The tribunal’s non-publication order remains in place, so LawNews has not identified the specific case that piqued Whyte’s interest to avoid further amateur sleuthing by other tech-savvy lawyers.
But the episode highlights the difficulty of ensuring name suppression is effective in the age of AI and algorithmically generated content, raising questions about how far the courts can go to enforce suppression orders and the obligations on the tech giants behind the technology.
AI ‘increasingly problematic’
Name suppression is designed to protect the fair trial rights of individuals and shield the privacy of vulnerable witnesses, such as children or victims of sexual offences.
It can also be granted in cases where publication would endanger safety, cause extreme hardship to a person or someone connected to them, and where it could bring suspicion on someone else, causing them undue hardship. Defence lawyers maintain it’s a high bar to cross.
Balanced against the restriction on naming people is the principle of open justice, allowing the public and media to scrutinise what happens in the court system.
As Judge Sanjay Patel writes on the District Court website: “A justice system that operated entirely behind closed doors would not have the public’s confidence.”
Human nature means that name suppression in high-profile cases inevitably generates speculation among the public about the person’s identity.
Data such as age, location and gender, plus descriptors such as “elite athlete” lead to what technology ethicist Dr Andrew Chen of the University of Auckland describes as “jigsaw identification” – combining fragments of information from multiple sources to identify a person.
While such speculative identification has always occurred, AI’s ability to collect and synthesise information from the deepest recesses of the internet has taken it to a new level.
“This is becoming increasingly problematic as search engines move towards an AI-first approach – for example, Google showing AI Overviews above search results and eventually moving to replace keyword search with ‘AI Mode’,” Chen told LawNews.
David Harvey
“An article or social media comment on an obscure website can be surfaced more easily to users looking for information about a case.”
Retired District Court Judge David Harvey said New Zealand’s name suppression regime was developed for the print era but needs updating to recognise the realities of a digital world.
“They envisaged a mainstream mass media that was largely centralised and which was obliged to comply with court directions. Failure to do so would attract sanctions,” he said.
“The distributed nature of online communications – not just social media – is paradigmatically different. An individual who breaches a suppression order can be held to account but not with as much ease as a centralised media organisation.”
‘They have to take responsibility’
While New Zealand media had a strong record of abiding by local name suppression orders, their overseas counterparts and the tech companies that control AI and social media are sometimes less compliant.
One prominent case of automatically generated content breaching suppression orders – although it predates AI as we know it today – came in 2018, when an automatically generated Google Trends email sent to New Zealanders named the man accused of murdering British backpacker Grace Millane.
The-then Justice Minister Andrew Little was furious, hauling NZ-based Google executives over the coals and expressing fears the breach could derail the trial of Millane’s accused killer (who was eventually convicted).
“If they choose to set up their algorithms and distribute news, they have to take responsibility for that,” Little said at the time. “We can’t just stand back and say ‘this is all too hard’… We have to protect the integrity of our court system.”
Last year, Google’s AI Overview reportedly identified defendants in several trials who had been granted non-publication orders, including former ACT Party president Tim Jago, who abandoned his bid for permanent name suppression in the wake of his conviction for sexually abusing teenaged boys.
‘Tipped towards suppression’
The case that has generated most debate in recent years concerns a member of a wealthy New Zealand family imprisoned for possessing child pornography. The courts permanently suppressed the man’s name, his family’s name, and the name of the company associated with the family.
Amid fierce public debate about the merits of granting the man name suppression, the X (formerly Twitter) AI platform Grok incorrectly identified him to users as businessman Wayne Wright Jr, based on social media speculation.
Another rich-lister, Mat Mowbray, was also falsely identified as the man, though on social media rather than via AI.
The National Business Review (NBR) led reporting on the case. The publication’s co-editors Hamish Mc Nichol and Will Mace said it highlighted the need to reconsider name suppression in the AI era.
“People will speculate and there is a danger of AI tools getting names wrong,” they said. “It puts even more importance on ensuring the balance between open justice and the need to protect fair trial rights etc is correct.”
McNichol and Mace called for a review of name suppression legislation to ensure it remains fit-for-purpose in the modern era. The pair fear a system they say is already “tipped towards suppression” will become even more restrictive as courts try to deal with the risks posed by AI.
“[The courts could react by] imposing more restrictions on what can be reported. We would be concerned that such a response would be against principles of open justice,” they said.
‘Media revealed too much’
Criminal defence barrister Emma Priest, who represented the convicted man, takes a different view. She says the case highlights the need to ensure that the descriptors used for name-suppressed individuals are tightly controlled so AI can’t make educated guesses about their identity.
“In this case, the descriptors reported were so narrow that AI misidentified two potential offenders,” she said. “This happened because the media revealed too much of the persons identifying particulars.
“The city a person lives in, together with their gender, age, job or other particulars can in some cases narrow the pool so much as to render the suppression redundant. In this case, the misidentification was quickly corrected and widely publicised.”
She added: “AI does not require a balance of the principle of open justice and the need to protect those who have suppression. Had AI breached name suppression in this case, it would be an offence under the Contempt of Court Act 2019. In this case, the published information was untrue and defamatory. It was quickly corrected.”
‘Fiendishly complex’
So how can the courts prevent AI-fuelled guesswork undermining name suppression orders without resorting to Draconian restrictions on what can be reported?
Harvey described it as “a fiendishly complex area” and Chen concedes there are no easy answers.
Harvey suggests judges should be required to provide more detailed reasons when granting suppression orders, saying the current regime gives then too much flexibility to justify them on the grounds of extreme hardship.
Rather than making indefinite non-publication orders, Harvey said the courts could also issue interim suppression orders, so an accused could advise family before they are identified publicly. This would help short-circuit the risk of AI speculation.
Chen said it was technically possible for AI companies to implement filters that block suppressed identities from their responses.
However, implementing such a solution would be challenging, not least because the AI companies are based outside New Zealand and there is no guarantee they would cooperate.
“There has been some suggestion of a name suppression register that could be shared digitally with certain publishers so that they can implement these filters automatically,” he said. “Although the existence of a single register with this sensitive information is also problematic.”
New Zealand has been down the road of blanket suppression orders before, when a law change in 1975 banned the publication of all defendants’ names until or unless they were convicted.
It lasted only 10 months and was repealed shortly after MP Gerald O’Brien was charged with indecency. While the case was swiftly dismissed by the courts, media were unable to publish O’Brien’s name while it was ongoing, leaving other male politicians scrambling to publicly declare it was not them.
‘Just a communications medium’
Priest points out that AI is capable of learning, meaning an AI prompt trying to identify her client now would generate a different answer to the one provided in 2025.
“Because of the misidentification in this case, the AI response is more cautious,” she said.
“The same search entered today will reveal a strongly worded response, telling people searching for the identify of a person with name suppression that the court has issued a permanent name suppression order and that it is a criminal offence to publish or share any details that could identify them.”
This proved correct when LawNews asked a well-known AI platform for its best guess about the man’s identity. The tool provided publicly available information such as his age when convicted last year and the fact that $50,000 in charitable donations had helped him gain a 3% discount of his sentence of two years and five months.
But it did not provide a name, saying: “I wouldn’t nominate a person based on family wealth, business connections or internet rumours, particularly when the courts have expressly protected the person’s identity.”
Pressed further, it said: “I don’t want to generate a shortlist of wealthy families that could falsely implicate innocent people” and stonewalled subsequent questions.
While acknowledging the complexity of the challenges thrown up by AI, Harvey noted that the justice system had previously faced similar issues posed by the emergence of new technology.
“I don’t share the fears of many about AI in the same way that I didn’t share the fears of computerisation when personal computing took hold, nor the fears that the internet would destroy civilisation as we know it – although it did usher in paradigmatic change in the way that we deal with information.
“After all, it is just a communications medium.”