The Edit is our weekly show, where Social Editor Grace Robinson quizzes Editor-in-Chief Ben Hanson on five of the most significant fashion and technology stories from the past seven days.
This edition covers the EU’s new ban on destroying unsold clothing and why it collides with the realities of range planning; the OpenAI models that broke out of their sandbox and into Hugging Face, and what that says about AI in cybersecurity; the rise of open-weights Chinese models and the pressure they put on closed-source incumbents; Apple, Klarna, Lefties and SHEIN’s IPO as competing answers to affordability; and what a Burnham government might realistically do for UK fashion retail.
This show is available in video format, embedded below, as well as audio, on Apple Podcasts and Spotify.
For deeper interviews with brand and technology leaders, our prestige interview show releases on Thursdays everywhere you find podcasts – and will be coming to YouTube soon.
Subscribers to our mailing list will receive notifications for both shows.
Grace Robinson: Welcome to The Edit from The Interline, the show where we run a quickfire analysis on our pick of the most important fashion and beauty technology stories from the last seven days. I’m Grace, the Social Editor, and I’m joined by Ben, the Editor-in-Chief. And together we have less than twenty-five minutes to give you our analysis on the stories that we think really matter.
Ben Hanson: Hello, Grace. Good to see you again. I am still travelling, hence the AirPods setup rather than the dedicated standalone mic. Travelling to a quieter place than I was for last week’s show, so hopefully people can pick up a difference in the audio.
Excited to get into the five stories that we’ve picked between the two of us. I think there’s some interesting crossover, and there’s some interesting divergence. We’ve a lot of ground to cover, so let’s roll.
New regulations restrict the destruction of unsold stock. Are the tools ready?
Grace Robinson: One of the biggest topics and one of the biggest news stories this week was the new ban on large companies destroying unsold new product. The measure has been introduced under the Ecodesign for Sustainable Products Regulation, and in summary it requires large businesses to keep products in use wherever possible. This could be by selling them, including through discount stores or alternative markets, donating them to charities or social enterprises, or preparing them for reuse through repair, refurbishment or remanufacturing. Now, from my understanding, a lot of big fashion brands do rely on being able to destroy products.
So I really wanted to know: do you think these new regulations will make big brands think about their buy more, both in terms of the actual products they’re planning but also the units behind them, so that they don’t over-forecast and over-produce products that they can’t sell? Or do you think big businesses will still find a way around this new rule?
Ben Hanson: I think over-forecasting and overproduction is a very, very difficult thing to solve using the methods that fashion traditionally relies on. Your typical lead time for developing a new product is something like nine months if you look at it from a category-agnostic, aggregated average point of view, which means that the plans you’re putting in place — whether they’re margin or financial plans, building slot plans, building assortments, whatever you’re determining to make — you’re making choices the best part of a year in advance of actual market conditions. Now, those choices are made at the product level and the style level and the colour level and everything else. You’re determining how much of something to make across all of those. It’s also made at the size level.
That’s the thing that people don’t often talk about when it comes to overproduction: size ranges and size ratios are quite out of whack for a lot of companies. What they end up with is a lot of the main sizing — mediums and so on — getting sold out quickly, and then they struggle to replenish them, and then they end up with a lot of excess inventory in smalls and larges and extra larges because, for whatever reason, they’ve overproduced in those. So there’s a big challenge in demand prediction and demand sensing, but also in size understanding. I had a fascinating podcast interview with Michaela Wessels, who’s the CEO of Style Arcade, a couple of months ago, and I’d recommend anybody who’s interested in the sizing side of this to give that one a listen.
Now, the reason that those products that are overproduced, overdeveloped and overstocked end up being destroyed rather than making their way into outlets or successive markdowns is usually brand protection. You’ll see this more at the luxury end of the market — Burberry and so on have, I believe, been criticised for this in the past. If you want to maintain the value of something, the perceived value of something, it cannot be seen to be oversupplied, and it cannot be seen to be discounted too much, because all of a sudden it no longer earns its premium price point.
What we’re talking about here is a big mess of size predictions, style forecasting, pricing, margins and planned markdowns. There’s a lot to unpick in terms of actually solving this. And to me, this strikes me as one of those areas where regulation is leading in a noble and a sensible way, but it’s quite separated from market reality. It’s quite separated from just how primitive and rudimentary a lot of the tools are that brands have to actually solve and address these things.
So I suspect in the short term, at least, you’ll see more people trying to find inventive ways to shift inventory to invisible channels — alternative markets and so on — than you will see them actually trying to take meaningful steps on addressing overproduction, overdevelopment and miscalibrated size ranges. Now, I need to refresh my memory on what the provisions of the ESPR are, but I believe there’s still a way for people to say, “I’m going to shift these to alternative markets. I’m going to move this excess inventory to where it doesn’t harm my brand perception domestically, and in the places that I care the most about.” That seems like an easier lever for a lot of people to pull than radical rethinks of how they forecast and build.
What it means that an AI model broke containment to beat a benchmark
Grace Robinson: This week we’ve seen a major story that, depending on your perspective, is either a piece of guerrilla marketing, or the cybersecurity event, or maybe the AI capability event of the year. I’m going to let you explain the full story, Ben, because it is quite messy and subjective.
But the reason I wanted to address this one was because we’ve seen so many cybersecurity and ransomware attacks in retail lately. There was a big wave of them in the UK last year, which you’ll remember — so M&S, Harrods and also the Co-op were affected, but major luxury conglomerates and mass-market groups were affected too, even though there’s still little out there to confirm who paid the ransoms.
So I wanted to know: what’s changed this week with AI that makes this story worth revisiting?
Ben Hanson: I’m going to work really hard to compress this one into a couple of minutes. I will say, I’m writing a news analysis of this at the moment, which we’ll publish on the day we record this. So by the time you listen to this, it will have been out for a couple of days. I’d recommend reading it.
Now, you’re correct that retail has seen a huge amount of cybersecurity and ransomware incidents. The ones you mentioned last year came from two named groups. Scattered Spider was one of them and ShinyHunters, I believe, was the other. They were very social-engineering-y hacks — people like to use “hacks” and “attacks” to mean the same thing. Those were much more of the kind of “let me find an individual working for a vendor, for either SaaS or something, for one of these brands, that gives me a wedge I can then drive into finding vulnerabilities and building exploitations in their systems”.
They still count as cybersecurity incidents because they still are expressed as an external party intruding into internal systems and taking a lot of sensitive data — in the case of retailers, because they have a lot of sensitive information about customer names and addresses and email addresses and payment methods and personalisation in particular. This is still there. It’s still people finding their way into systems. What happened this week was really weird to talk about and really weird to analyse.
The main catalyst is that OpenAI announced that they were testing a currently unreleased model alongside a released model. The one that we know about is 5.6 Sol; the unreleased one is presumably GPT-6. And they were testing those in what’s referred to as a sandbox — so a closed environment that is, for all intents and purposes, air-gapped from the wider internet. And the point of that evaluation was for those models to get scored on a benchmark called ExploitGym. The point of that benchmark is to set the model a task and say: here is a known vulnerability in one or more systems. It could be JavaScript, it could be anything else. Your task is to build an exploit on this, and that’s the sole way that success will be determined.
What happened was that those models apparently found a vulnerability in the sandbox environment itself, so they were able to connect to the wider internet. When they connected to the wider internet, they determined that the most efficacious way to solve this benchmark was to go and directly find the answers. They inferred that the answers might be housed on Hugging Face — Hugging Face being the big open-source, open-weights AI community. So the models performed a combination of probing, intrusion testing and social engineering to obtain login credentials that allowed them to get into Hugging Face systems, and then to move laterally across those to find the answers to this benchmark and then score full marks.
All of that sounds weird and super scary to anybody who’s not deeply immersed in AI. For me, the big takeaways from this are: yes, there is an emerging frontier in AI cybersecurity, and a lot of what I’ve just described is wild to think about. The key thing for fashion to remember, though, is that AI models do what they’re told. So what you have here is an instance of an AI model being set a very specific goal and going out and pursuing it through a lot of different adjacent and lateral means.
The real threat to fashion — this is there, and I would recommend that people invest in shoring up their security and looking at alternative models as a way of doing this — but the real threat is still the social engineering stuff. The real threat is still people looking to extract data and then extract ransoms from companies, and they’re tending to drive those wedges using much more traditional, person-to-person social engineering. But I will say AI capability in general in cybersecurity is on a massive upward trajectory, and I do think it’s something fashion needs to keep tabs on.
Open weights and Chinese AI are offering near-frontier capability outside the closed ecosystems.
Grace Robinson: I think the next story is actually tied to the one that we just tackled, because it involves some of the same key players — so OpenAI and Hugging Face, like you mentioned. But it also extends out to the open-weights AI community, and especially to Chinese AI labs that produce model series like Kimi, GLM, MiniMax, DeepSeek and so on. I feel that we’ve been hearing a lot about the threat of AI developed outside the US for a while now, but it’s back in the news in a big way. So tell me why, and tell me how this is going to affect fashion.
Ben Hanson: Yes. You and I have talked before about the Anthropic Fable 5 mythos situation, where the US government stipulated that that model had to be withdrawn from serving and distribution and had to undergo a lengthy process. The follow-on to that, which you and I haven’t talked about, is that the US has now instituted a policy where all frontier AI models need to undergo a thirty-day pre-inspection process from government officials. Prior to being made publicly available, they need to be made quietly available to the American government and to, I believe, a preselected list of US enterprises. So we have quietly already found ourselves in a situation where American companies and American providers are at least thirty days ahead of the AI that is broadly available outside the US, or to your general-purpose US enterprise.
Now, what’s been particularly interesting this week is that in the story we just talked about — the Hugging Face intrusion by the OpenAI models that broke containment — the CEO, or the CSO, I forget, of Hugging Face said that they initially tried to solve what they thought was an attack coming from somebody outside. They had no idea at the time that it was OpenAI. They initially tried to solve it by using the models that are available to every company in the US, the closed-source ones. So your ChatGPT, GPT series, your Claude series and so on. Then they quickly hit the guardrails and the classifiers of those, which is what the US government has required to be put in place for those models to hit broad availability.
If you try to talk to Fable 5 about cybersecurity matters, it will punt, and it will pass that to a lower-capability model. That’s just what’s built into it. So what ended up happening was that the team at Hugging Face turned to open-weights Chinese models — I think it was GLM 5.2 — to solve the intrusion as it was happening at the time. Now, this also coincides with the release of Kimi K3, I believe it is. MiniMax M3 has been around for a while. DeepSeek V4 Pro. There’s a whole bunch happening in Chinese AI, which is primarily open weights because Xi Jinping has made that a priority of Chinese AI development.
Now, open weights in this case is not the exact same thing as open source, but it does mean that you can take these models and run them yourself if you have the hardware to do it, or you can partner with cloud and inference providers to do it for you and actually serve those models. They’re not locked down in the way that closed-source models are. And there’s a big risk here, as the closed-source AI labs see it, that near-frontier capabilities are becoming available outside the closed ecosystems of Claude and ChatGPT and so on.
Now, in a sense, this isn’t really news, because if you refer back to an interview with Francesco Bottigliero from Brunello Cucinelli last year or earlier this year — I forget — he was talking about their use of open-weights and open-source models for a number of the things that they do. What’s interesting is that if you’re a fashion technology company that is currently very much a ChatGPT house or a Claude house, you may find sooner rather than later that you can get similar capabilities, open weights or open source, for potentially a fraction of the cost. And this is a very interesting emerging frontier. It hasn’t quite resulted in a huge amount of stock value loss in the way that I think people expected, or in the way that happened with the first DeepSeek V1.
But Chinese AI, open-weights AI, is moving really quickly, and companies that are reliant on closed-source US AI should be keeping their eyes open.
Affordability: leasing iPhones, Lefties, and real competition in the value market.
Grace Robinson: The next story is actually something that both you and I flagged this week, and it’s the news that Apple has teamed up with Klarna to offer a new lease-to-own programme for devices like iPhones, iPads and Macs. Now, last week we talked about some of the new rules here in the UK about fintech companies that spread the cost of purchases, but this feels like something a bit different — especially when we weigh it up with two other stories that you sent me. One is about Inditex launching a value-first brand in the UK for the first time, and the other is SHEIN targeting an IPO that’s way down from what it maybe expected to be.
So I wanted to know: why are these stories related, and why are we talking about leasing iPhones on a fashion technology show?
Ben Hanson: I think we’re talking about leasing iPhones on a fashion technology show for two reasons. One is that, as anybody who’s listened to these shows week over week will know from our previous conversations, I’m not keen on technology being deployed in service of encouraging people to buy things they can’t afford. And this is very much a manifestation of that. The regulations that are being put in place, just as a quick refresher, apply the same standards to Klarna and similar BNPL providers that are applied to traditional credit lending, to avoid usury situations. At least here in the UK, that is the objective.
And you and I have talked before about what I see as a risk, which is that the more AI acts as an intermediary in product discovery and buying decisions, the more people end up having persuasive interactions through text and voice that then lead them to transactional interactions that have their impact softened and spread by technology providers who are effectively doing relatively high-APR lending, and putting people in uncomfortable positions.
Now, the interesting part to me is that it’s not that you can lease an iPhone — fine. You’ve always been able to get on an iPhone upgrade programme; there have always been multiple ways. A lot of technology is based on upfront subsidies and things. It’s more that it’s very clear the value market is where the competition is. If you provide a premium good, your way of getting across that and trying to hedge against that is to make your premium good affordable by spreading the cost of it.
The value market, though, has been SHEIN and Temu’s to own for the last couple of years, right? France, the EU, a lot of places have tried to regulate the model away; they’ve tried to close loopholes and things. But the reason that value brands have been performing exceptionally well is that people don’t have the money to buy more expensive things. People like cheaper stuff. There’s a whole behavioural situation there. Now, the fact that SHEIN is looking to debut its initial public offering at something like half of the projected valuation from 2022 is almost a good thing to me, because it shows how much competition there actually is in that value space.
And a prime example of that competition is Zara — sorry, Inditex, the parent company of Zara — launching its Lefties brand, which really emphasises value. And it emphasises value at the product level, but also at the operations and store level. So there’s a lot of RFID tagging, self-checkout, robot sorting. There’s a lot of technology deployed in service of decreasing the operating costs of running these kinds of brand and retail operations, to then drive the price for the consumer down.
To me, I think there’s a huge amount of innovation that’s coming in a place that people don’t expect. I think people look at this fintech stuff and they think, oh, that’s the future — we spread costs, we try to make these premium goods as affordable as we can, and that’s how we respond to the cost-of-living crisis. To me, it’s far more interesting to look at the value space and to say: well, people like buying cheap things for a reason, and one of those reasons is that they don’t have the discretionary income to buy more expensive things. What if we innovate in how we deliver cheap goods, in order to actually create greater competition in the value market as well? That might be a leap from why we’re talking about iPhones, but that’s what I was driving at.
What can a Burnham government realistically do for fashion?
Grace Robinson: Now, as you’ll know, the UK has a new Prime Minister, Andy Burnham, and with him now in power there have already been discussions about the country’s future relationship with the European Union, including some suggestions that closer ties could be pursued over time.
Now, for those who don’t know, the 2016 Brexit referendum did have a significant impact on trade between UK and EU fashion businesses. For example, the wholesale platform JOOR’s data shows that EU retailers reduced UK brands from 17% of their total buy in 2017 to under 7% in 2025. Over the same period, EU brands more than tripled their gross merchandise value, while the UK’s share of that transaction volume fell from 12% to under 4%.
So with the new government in place, a lot of fashion industry leaders are calling for policies that will help restore the UK, and also specifically London, to the position of being a leading global shopping destination. For instance, the Burberry CEO, Joshua Schulman, has urged the government to reinstate tax-free shopping for international tourists, and similarly the BRC’s Chief Executive, Helen Dickinson, has argued that cumulative costs and fragmented policy decisions are actually holding the fashion industry back.
So my question to you, Ben, is: do you think Burnham will make fashion retail a priority, and what should large fashion businesses be focusing on during this time of political and economic change?
Ben Hanson: As you and I have talked about before, you’re based in London and I’m based in Manchester. Andy Burnham, former mayor of Greater Manchester, is now exported down to London to be Prime Minister. So I technically do have something of a horse in this race, in that I live in a city that has benefited quite a lot from the leadership of Andy Burnham — not exclusively. Manchester is the fastest-growing city region in the UK, and I think it’s up there in Europe as well. So there are a lot of policies instituted here that I think he’s looking to extend across the UK.
All that aside, I think you have to remember, and I certainly have to remind myself, that every successive Prime Minister that we’ve had — and we’ve had a lot of them in the last five years or so — they’re bound by what it’s actually feasible to accomplish, and they’re bound by a lot of the policy objectives set by their predecessor, who was the one who was actually elected. So in this case, Keir Starmer, who won a general election. I think the British Retail Consortium, the BRC, is right to be campaigning this way. I think Burberry are right to be campaigning this way. I question what kind of power Andy Burnham actually has to take the sort of action that they’re looking for here.
I think the tourist tax thing is within his remit. I think broadly catalysing retail and putting retail on a better footing is harder and bigger. It’s a more complicated task. For people outside the UK, the biggest announcement to have come from the administration for hospitality and retail has been focused on pubs, live music venues and, I believe, bars, but not restaurants, and it’s come in the form of business rate relief for those sectors. It’s something like 2%. It’s marginal, but it’s been potentially valuable for those sectors. What ends up happening inevitably is that you then have restaurants who complain about that side of things — why could these business rate reliefs not be applied across the board?
I think we might see some preferential treatment of fashion a little bit further down the line, on the basis of wanting to restore the UK’s place as a shopping destination — that’s one. And I do think the Manchester connection has some bearing here, because as a seat for the textile industry and everything else, and also being a place where a lot of brands and tech companies are currently based, I think Andy Burnham will be aware of the export side of things when it comes to fashion.
Now, nobody’s going to unwind Brexit. I think there’s no political capital to do that. There’s very little will to do that. I do think we’ll start to see more in the way of export posture and positioning from the UK. How receptive the EU will be to it, how receptive your big luxury stockists in France — that I assume are where some of this JOOR data is coming from, Galeries Lafayette and so on — are they suddenly going to want to bring in more British luxury brands, in the protectionist mode that France in particular is in at the moment?
I could be wrong about all of this. I’m not a politician or a political analyst, but it’s an interesting space to watch at the very least.