Huffer's forensic report doesn't settle model likeness claims, say experts
Saturday, 11 July 2026
A month on from public claims Huffer used the likeness of real models to create AI-generated marketing images, the brand says a forensic report has exonerated it. But experts tell Stewart Sowman-Lund the debate may not yet be settled.
Artificial intelligence experts say a forensic investigation commissioned by fashion brand Huffer does not conclusively rule out claims the company used computer generation to mimic the likenesses of real models, despite the company saying it had been cleared.
And one academic has also questioned whether a digital forensics specialist was the most appropriate person to lead the investigation.
But Huffer has pushed back, standing by the findings of the investigation, which it says found “no evidence” it had reproduced or generated the identity or likeness of “any model”.
Read more:
Forensic probe finds no evidence of AI model-copying claims: Huffer
Huffer withdraws legal letter against model who challenged AI images
Huffer made public a summary of findings from the investigation, saying it “established that no facial images of any of the individuals whose likenesses were alleged to have been used were ever provided to, uploaded into, or used by the AI image generation process”.
It followed a month of pressure on Huffer to address claims it had used AI to create computer-generated images that looked like models the brand had previously worked with.
While the company had already accepted it had used an “AI design tool”, it has maintained that the likenesses of real models were not mimicked.
The forensic investigation was conducted by forensic technology and cyber security expert Campbell McKenzie, the managing director of IRS.
According to the report’s summary, released by Huffer through an external publicity firm, the investigation “concluded it could not identify any prompt, image reference, workflow record or output, indicating that no individual's facial likeness was used as an input or reference when generating the images”.
In its investigation, IRS examined Huffer’s workflow and prompt history - which is how users instruct the technology - and found that the prompts used to generate the images did not “reference any specific model, individual, or previous Huffer modeling work”.
The report concludes that these findings are “supported by evidence to a standard capable of standing up in court”.
Media were provided a two-page summary of the report, with a spokesperson saying the full investigation would only be made available subject to “appropriate confidentiality arrangements” as it contained “legally privileged and commercially sensitive material”. Nobody from Huffer was available to be interviewed.
The Post contacted McKenzie, but he would only comment via questions submitted to public relations firm Pead PR that Huffer has engaged to lead its communications in the ongoing saga.
In response, he said the findings of the review were clear. “It did not identify any evidence that the faces shown in the images were generated from, or by direct reference to, facial images of any of the individuals whose likenesses were alleged to have been used.”
McKenzie was “confident” that there were “no blind spots” in the investigation.
The Post provided Huffer’s summary findings to a number of AI experts, who said that while it may show the fashion brand did not intend to recreate the image of real people, that doesn’t rule out that it occurred.
Dr Andrew Lensen, a senior lecturer in artificial intelligence at Victoria University, said there were unanswered questions.
“For example, the report states that product images supplied for use in the AI workflow had models' faces deliberately redacted beforehand. This doesn’t guarantee that the AI system won’t be able to infer the model’s face anyway -an AI system that was trained on Huffer’s (publicly available) images could have learned that a specific model tended to wear certain clothes, have certain poses, or other non-facial characteristics,” he said.
“If so, when an image without the model’s face is given to the AI system, it is plausible that the AI system would create new images with faces that resemble the model whose face was redacted from the input image.”
Lensen said that while he was not saying the AI system definitively did recreate a model’s likeness, “I don’t think the investigation quite reaches the threshold of proving that they didn’t, based on the summary provided”.
He also questioned why Huffer commissioned a forensic technology and cyber security company to carry out its review, noting that an AI expert may have been “more conservative”.
McKenzie said his work included digital forensic investigations, cyber incident response, electronic discovery and expert evidence in relation to digital systems and electronic records which included “experience in the use and application of artificial intelligence (AI) image, video, and audio generators”.
Huffer, too, defended McKenzie’s expertise. Said a spokesperson: “It’s a bit like investigating a suspected art forgery. You don't ask an art historian to explain the history of painting; you ask a forensic examiner to determine whether a particular painting was copied from another.”
Another AI expert, Dr Andrew Chen, an honorary research associate at the University of Auckland, said it seemed as though Huffer did not intend to replicate real peoples’ likeness, but it possibly still could have happened.
That’s because Huffer had a lot of images with these models publicly available, “and those images have likely been collected by the large AI companies and used in their training data”.
He continued: “If Huffer has then said, ‘Here are some clothes that we would like you to put on some people’, those AI models will say, ‘Well, this is a similar style to other Huffer clothing that we've looked at before, and this is what people who wear Huffer clothing look like’, and so created images that reflect that likeness.”
Chen said the fashion label could have noted the similarities between the AI images and real people before choosing to publish them.
“Could Huffer have looked at those images and said, ‘Oh, that looks like someone we know, maybe we should change the face a bit and make it look like someone else, or make it not look like that person?’ Maybe. But these are all real sort of grey ethical areas.”
But Huffer said that “confuses the technology with the way it was used”.
“The question of what data may have been used to train third-party foundation models is a broader industry issue and was not the allegation made against Huffer.”
Auckland University consumer law expert Professor Alex Sims said the summary of Huffer’s report did not say that images of the individuals were part of the AI’s training data, describing this data as “effectively a black box”.
“As images of Huffer’s garments were provided to the AI, it is conceivable that the AI made a match between Huffer’s logo and other earlier photographs bearing that logo, which also featured the individuals.”
A technology professional with knowledge of generative AI systems, speaking anonymously to protect their professional relationships, also believed that the investigation concluding there was “no evidence” that model’s likenesses were used did not mean it had not happened.
“The report only looked for facial copying in local files and prompt logs. It openly admits that real shoot photograph shaped other parts of the generated images,” they said.
They said the investigation also appeared not to consider that bodies may have mimicked, as “the report drew a very specific line around faces and ignored everything else“.
They added: “The report finding no evidence means the places they looked didn’t contain evidence, not that proof doesn’t exist … the original complaints were always about the whole picture, the body, the post, the composition, not just the face.”
Huffer said its report did not ignore those issues. “It considered the complete AI workflow and the source material used in creating the images,” a spokesperson said.
“As the report explains, non-facial elements of that imagery may have influenced aspects of the final images. However, it found no evidence that this involved reproducing or generating the identity or likeness of any model.”
Lensen believed “the bigger story” was that Huffer was not “reading the room” when it came to how its audience viewed AI.
“Many people, including their target demographic, are very concerned about AI’s impact on the creative and modelling sectors,” he said.
“In addition to AI job losses and environmental harms, there is a risk that AI images do not realistically ‘model’ clothing, e.g. by presenting it on a model who it would not fit/suit in reality.”
Huffer did not “accept that characterisation” and said that “AI Design is used to enhance the creative work of our large creative, product, and development team. It is not a substitute for the real people who bring our product and collections to life”.