Fake fashion hits a record in Europe as AI arms both counterfeiters and detectors

European enforcement authorities detained more than 196 million counterfeit articles in 2025, with an estimated retail value of EUR 6.3 billion, according to the annual report by the EU Intellectual Property Office (EUIPO) and the European Commission published on 22 October 2026. Volumes were up 67% on 2024 and value almost 63%, a second consecutive all-time record. One detail should worry every fashion brand: labels, tags and stickers jumped from 2.11% of items detained inside the EU market in 2024 to 13.41% in 2025. The fakes are increasingly finished near the customer, with a brand's identity stitched on at the last step.

That record sits against a much larger estimate. The OECD and EUIPO put global trade in fakes at USD 467 billion in 2021, the latest year with comprehensive data, and EU imports of counterfeits at EUR 117 billion, or 4.7% of all imports into the bloc. The direct answer to the question behind this dossier is that artificial intelligence now works on both sides of the market. Sellers use it to produce convincing photos, listings and whole webshops at speed; platforms, brands and resellers use it to scan billions of listing changes, recognise patterns in seller behaviour and inspect the surface of a handbag through a phone camera.

It matters now because the channel has shifted. Counterfeits used to arrive in containers; today they increasingly arrive in small parcels ordered online, found through social media ads or AI-assisted search. The question for brand owners is no longer whether to use AI against fakes, but which claims about its accuracy deserve to be believed.

How big is the global trade in fake fashion?

The OECD and EUIPO estimate, published in May 2025, is based on customs seizure data and is the most widely cited global measure. It found that clothing, footwear and leather goods remain the most seized counterfeit categories, accounting for 62% of seizures, according to a summary in Global Finance magazine. China was the source of 45% of all reported seizures in 2021, and the EUIPO summary notes that the EU mainly imports counterfeits from China and Hong Kong. The report also underlined the role of free trade zones, which with fewer controls favour the trade.

Two further findings point to the online shift. In Operation Stop III, an action involving 111 customs administrations, 71% of cases involved parcels ordered over the internet. And goods seized at US borders for intellectual property violations more than doubled between 2020 and 2024. "Illicit trade threatens public safety, undermines intellectual property rights, and hampers economic growth," said Mathias Cormann, the OECD Secretary-General.

The headline estimate should be read with care. It is a modelled figure extrapolated from seizures, not a count of fake goods sold, and its 2021 base year predates the rise of the newest marketplace platforms. It is the best available order of magnitude, not a precise measure.

Why did EU counterfeit detentions hit a record in 2025?

The 2025 EU report shows two very different stories at the border and inside the single market. Customs at the external border detained 42.1 million articles, up 114%, with an estimated value of almost EUR 2.0 billion, the highest since Regulation (EU) 608/2013 on customs enforcement took effect. Much of the jump came from 16 million items of other tobacco products. Inside the EU market, police and market authorities detained more than 175 million articles worth EUR 4.6 billion, up nearly 71% on 2024. Italy alone accounted for 80.21% of articles detained in the internal market, which shows how dependent the statistics are on the intensity of enforcement in a single country.

The composition of internal-market detentions explains why fashion should read beyond the totals. Recorded CDs and DVDs, including software, made up 32.78% of articles detained, down from 40.02% in 2024. Toys accounted for 19.47%. Labels, tags and stickers rose to 13.41% from 2.11%, and games to 4.32% from 1.29%. Loose brand labels are the raw material of fake apparel: they let counterfeiters import or produce unbranded garments and apply the logo locally, out of sight of border controls.

At the border, the origin picture is familiar. China was the provenance of 40.77% of articles detained by EU customs in 2025, Türkiye of 18.63% and Hong Kong of 5.70%; all other countries together accounted for the remaining 34.90%. Channels have fragmented: express courier was the largest source of registered cases with 45,094, and the one with the largest increase over 2024, followed by postal traffic with 26,453 cases and air cargo with 14,836. In 84.69% of cases the goods were destroyed.

A record is not the same as a surge in fakes. Detentions measure enforcement as much as supply, and the 2025 figures depend heavily on one tobacco category at the border and one country inside the market. The previous year illustrates the point: in 2024 the EU detained 112 million counterfeit items worth EUR 3.8 billion, fewer items than the exceptional 2023 but a record value, which the Commission attributed to higher unit prices. Clothing ranked among the main categories that year, and China, Türkiye and the United Arab Emirates were the top three origins at the border.

How is AI changing the way counterfeits are sold?

The clearest evidence comes from consumer research, and it should be read as such. In May 2025 Red Points, a brand protection company, and the market research firm OnePoll published a survey of 2,000 US consumers who had bought counterfeit goods in the previous two years, plus input from 1,000 brands. Some 46% said they had bought counterfeits accidentally, 39% knowingly and 15% both ways. Among accidental buyers, 57% were drawn in by convincing product photos and 47% by listings that imitated real brands. Some 61% did not realise a product was fake until it arrived.

The survey links this to new tools on both sides of the purchase. 28% of consumers who bought knock-offs online said they had used AI tools in their search, and 34% of fakes were bought on fake websites, often found through social media ads or posts. Red Points said it detected 4.3 million counterfeit infringements online in 2024, up 15%, and that social media ads redirecting to infringing websites rose 179% between 2023 and 2024. It projected that fraudulent websites enabled by generative AI would rise 70% in 2025. That projection is the company's own and was not independently verified. "We're now seeing AI shape both the threat and the solution," said Laura Urquizu, Red Points' chief executive.

The mechanism is easy to understand. Generative tools reduce the cost of producing a believable storefront, product photography and copy in many languages to almost nothing. A counterfeiter who once needed a photographer and a translator can now launch, abandon and relaunch shops faster than a legal team can file takedown notices. That shifts the advantage to whoever detects patterns earliest, which is where the platforms' own AI comes in.

What are platforms and brands doing to take fakes down?

Amazon published its first Trustworthy Shopping Experience Report in April 2026, covering 2025. The company said it identified, seized and disposed of more than 15 million counterfeit products worldwide during the year. Its Counterfeit Crimes Unit has pursued more than 32,000 bad actors through litigation and criminal referrals in 14 countries since 2020, and collaboration with Chinese law enforcement and brands led to more than 70 local raids in 2025. Amazon described multimodal AI systems that analyse text, images, seller behaviour and supply chain patterns together and scan billions of attempted product detail page changes a day. In one example it gave, infringing listings for a viral new brand were blocked eight days before the brand owner shared its intellectual property with Amazon.

These are company claims. The report does not disclose how many counterfeit listings went live before removal, how many notices brands filed, or what share of fakes AI caught compared with human reviewers, so outsiders cannot calculate a detection rate. "Our goal is to protect the store for customers, brands, and sellers alike," wrote Rohan Oommen, Amazon's vice-president of worldwide customer and partner trust.

Brands run their own programmes, and some publish unusually detailed numbers. Moncler reported that in 2024 its anti-counterfeiting work led to more than 2,800 seizure cases worldwide, removing about 150,000 finished products and 177,000 counterfeit branded items such as logos and labels. Online, it reported the removal of more than 92,000 auction listings, the blocking of nearly 300 websites, the removal of almost 32,000 links from major search engines and around 203,000 social media posts, ads and accounts. For Stone Island, owned by the same group, it reported more than 38,000 listings removed, more than 130 sites closed and nearly 99,000 social posts, accounts and ads taken down.

The ratio is the revealing part. For every physical seizure case at Moncler there were more than a hundred online actions, and social media takedowns outnumbered auction listings by more than two to one. The fight against fakes has become largely a fight against listings, ads and accounts, which is exactly the terrain where generative AI lowers the cost for the counterfeiter.

Can AI tell a real handbag from a fake?

Entrupy, founded in 2012 and developed with research at New York University, is the best-known AI authentication service for luxury goods. Users photograph an item through its app; the system compares highly magnified images against what it describes as a database of millions of records of authentic and counterfeit products and returns a verdict of authentic or unidentified. The company says at least one expert authenticator reviews every result and that each authentication is financially backed for a year.

Its accuracy claims are high but inconsistent. The current website promises "nearly 100% accuracy" in one place, a "99.86 percent accuracy rate" in its FAQ and "over 99 percent" in another answer. In December 2023 FashionUnited reported a claimed accuracy rate of 99.1%. We found no independent, published test of these figures, and Entrupy states that it is not affiliated with the brands it covers and that designers may not honour its certificates. A claim of 99% also says nothing on its own about how errors split between fakes passed as genuine and genuine items flagged as fakes, which matter very differently to a reseller.

Resale platforms show the limits of the technology in practice. The RealReal says in its annual report for 2025 that it uses artificial intelligence to make initial assessments of authenticity, backed by gemologists, horologists and brand experts who inspect thousands of items a day in four authentication centres. It also warns investors that it "cannot be certain that every counterfeit item will be identified", and that its ability to identify counterfeits may decrease when it does not take physical possession of goods. The caution has history: in 2021 the company agreed to pay USD 11 million to settle an investor class action alleging that its authentication process fell short of its description, Retail Dive reported.

Will digital product passports make authentication easier?

The alternative to judging a product by its appearance is to give it a verifiable identity at birth. The Aura Blockchain Consortium, co-founded by LVMH, Prada Group, Cartier (Richemont) and OTB, has passed 50 million registered products and has around 40 members across fashion, jewellery, watches, cars and spirits, Journal du Luxe reported. Its digital product passports, usually read through a QR code or NFC chip, are designed to be tamper-proof and can issue certificates of ownership that travel with an item into the resale market. Brands named as users include Loro Piana and Rimowa, which puts an NFC chip in its suitcases.

Individual brands have used similar tools for years. Moncler has given products a unique alphanumeric code and an NFC tag since 2021, which consumers can verify online. Stone Island has used Certilogo authentication since spring and summer 2014. The logic is to move the burden of proof from detecting a fake to confirming a genuine item.

The limitation is coverage. A passport protects only products that carry one, and a buyer who never scans a tag gains nothing. Counterfeiters can also copy the visible code or chip housing, so the security depends on the cryptography behind the scan and on consumers knowing what a genuine result looks like. Passports are strongest in resale and after-sales service, where the owner has an incentive to prove authenticity.

What are the limits of the numbers and the detectors?

Every data source in this field has a bias. Seizure statistics measure where enforcement is strongest: Italy's share of EU internal-market detentions says as much about its police as about its market. The OECD and EUIPO figure is a modelled estimate built on seizures, and its latest base year is 2021. Platform reports present what companies choose to count, and vendor surveys, such as Red Points', are published by firms that sell brand protection services.

Detection claims carry the same caveat. None of the AI systems reviewed for this article publishes independent, audited error rates broken down by false positives and false negatives. A counterfeiter who knows which features a model inspects can adapt; a model trained on yesterday's fakes may miss tomorrow's. And aggressive automated takedowns can hit legitimate resellers and parallel importers, a cost that rarely appears in brand protection statistics.

There is also an economic counter-argument. Many buyers choose fakes knowingly: 39% of the counterfeit buyers in the Red Points survey said they bought on purpose. Detection technology does little against deliberate demand, which responds more to price, availability and social attitudes than to algorithms.

What should fashion brands do now?

1. Track labels and trims, not only garments The jump in detained labels and tags suggests fakes are assembled near the market; work with customs and suppliers on control of branded trims.
2. Demand error rates from vendors Ask any authentication or monitoring provider for false positive and false negative rates, test set size and whether results were independently checked.
3. Register intellectual property early Amazon's example of blocking listings eight days before a brand shared its rights shows the value of getting trademarks and designs into platform programmes before a product goes viral.
4. Prioritise ads and social accounts Takedowns of posts, ads and accounts now far outnumber physical seizures; allocate monitoring to the channels where buyers are deceived.
5. Give products a verifiable identity Tags, codes or digital passports at source move the burden from spotting fakes to confirming genuine items, especially in resale.
6. Measure outcomes, not activity Report how often customers encounter fakes and how fast they disappear, not only how many listings were removed.