play · digital · health
same toy, new seller
In the EU and the UK, takedowns of dangerous toy listings reach only "identical" copies; look-alikes from other sellers often stay on sale, and those bought and tested have failed safety tests
Source: DGCCRF Open ↗source: primaryLast updated: 2026-11-17
Problem statement
Many toys for babies and toddlers are sold in Europe by third-party sellers on large online marketplaces, often sellers based outside the EU. Official test buys keep finding dangerous ones. In 2025, France's consumer-protection authority (DGCCRF) bought toys on seven foreign marketplaces without revealing who it was. Of 316 games and toys its laboratory analysed, 78% were non-compliant and 42% were dangerous. When a toy is found unsafe, the remedy works listing by listing: the authority notifies the platform, the platform removes that listing, and the alert goes onto the EU's Safety Gate system. The law lets an authority order the removal of "all identical content" too, but only where the search is "limited to the information identified in the order" and needs no "independent assessment" by the platform (General Product Safety Regulation, Article 22(5)). But the same toy often comes back, or never left, under another seller, another manufacturer and another named "EU responsible person". In 2025 the industry body Toy Industries of Europe (TIE) bought 10 toys that looked identical to toys that had failed its 2024 tests, on the same platforms and sometimes from the same seller, and all 10 failed again. It also bought a baby toy that matched a Safety Gate alert whose recorded outcome was "Removal of this product listing by the online marketplace". It found that toy, or very similar ones, still widely sold: three samples from three platforms, each with a different seller, manufacturer and EU responsible person, and all three had the defect named in the alert. TIE is the toy industry's trade body, so these are its own results. Independent checks point the same way. In the EU's 2025 sweep of childcare-article offers, national authorities found 10 cases where a removed product or listing reappeared. In the UK, the consumer group Which? had 15 look-alikes of flagged products lab-tested (13 electrical goods and 2 toys), and all 15 failed UK rules. The unsolved problem is how to act on look-alikes. No method has been agreed for deciding, from a listing and without lab-testing every copy, when a toy that looks like a flagged one should be treated as the flagged one.
Why this matters
Many of the toys involved are sold for children under three, and the hazards are the old, well-understood ones. Toys were the second most often notified product type on Safety Gate in 2025 (16% of 4,671 alerts), and the Commission says choking risk falls mainly on small children, from small parts in toys or clothing (Safety Gate 2025 report). In TIE's 70 toys, the most common failure was small parts present or released in testing (26 toys), followed by easy access to batteries (7 toys, three of them with lithium coin cells), plus strong small magnets, suction cups that came off, and plastic bags too thin. TIE says "basic laboratory testing would have identified all the issues" and that many toys failed on design, not because samples varied. In France, the DGCCRF reports that 260 product references have already been the subject of an alert, and that more than 100,000 units, 57,000 of them toys, have been withdrawn from sale. The scale online is large too. Euroconsumers reports Temu at 115.7 million and Shein at about 145.7 million monthly users in 2025 (from the platforms' own transparency reports). In the UK, Which? found that the average gap between a regulator's warning and its finding a matching product still on sale was 175 days.
What’s been tried and why it hasn’t worked
Takedowns of single listings. Under the EU's Digital Services Act (DSA) and the General Product Safety Regulation (GPSR), platforms must remove products reported as dangerous. The DGCCRF summary adds that they must also run random checks of listings against products already flagged. An authority's removal order may extend to "all identical content", but only within the information in the order, without "an independent assessment" by the platform, and where "reliable automated tools" can do the search (GPSR, Article 22(5)). On our reading, a look-alike under a new name, seller and photo set falls outside that wording. In the first coordinated EU "sweep" (1 April to 15 May 2025), national authorities checked 1,741 childcare-article offers on 47 marketplaces and sent 252 orders. They found 16 offers of a dangerous product already notified to Safety Gate, and 10 cases where a removed product or listing reappeared. When the consumer group Testachats/Testaankoop, a Euroconsumers member, reported faulty products to Temu and Shein, Shein removed the reported items within days. Temu answered only after several weeks, saying it had removed the items or was removing them. But "neither platform removed any similar products that could pose the same risks", and many comparable items stayed on sale. In 27 children's products from each platform, Temu had 26 non-compliant and Shein had none that met the test standards.
The "EU responsible person". EU rules require a named person in the EU who answers for the product. In the sweep, 32% of listings did not give that person's details in full, and some authorities found named persons who were wrong, did not exist, or did not know they had been named. In TIE's sample, 8 EU responsible persons were named on 38 of the toys tested, all of which had safety issues, and "it does not look like they check compliance of the toys." TIE adds that platforms are not obliged to check that the address is real or that any agreement exists.
The platforms' voluntary pledge. Under the Commission's Product Safety Pledge+, signing marketplaces commit to "make efforts to complete the identification of relevant (identical) product listings within 3 working days" after a notice or a Safety Gate alert, "with additional flexibility for products that do not carry unique identifiers", and to take "measures aimed at preventing the reappearance of dangerous product listings already removed". The Commission counts 12 signatories (Safety Gate 2025 report). The pledge does not define "identical". The pledge text, dated 28 March 2023, lists 11 signatories: Allegro, AliExpress, Amazon, bol.com, Cdiscount, eBay, EMAG, Etsy, Joom, Rakuten France and Wish. Four of the seven platforms TIE bought from in 2025 are on that list (AliExpress, Amazon, Cdiscount and Joom); Fruugo, Shein and Temu are not. The sources read do not name the twelfth signatory.
Automated searching for flagged products (already built). The Commission's eSurveillance web crawler "automatically lists products that have been notified on Safety Gate but are still sold or reappear in web shops and online marketplaces" (Safety Gate 2024 report). In 2025 more than 1.6 million URLs were inspected and more than 20,800 "of these URLs" contained dangerous products (the report prints the second figure under the heading of its newer "Proactive web crawler", so which tool found them is not clear). The eSurveillance crawler generates contact links for notifying the marketplace, which "must react within three working days". The same year the Commission launched an AI-based "Proactive web crawler" that searches marketplaces for "potentially unsafe or non-compliant products" and early safety trends (Safety Gate 2025 report). The 2025 report gives no rule or method for deciding when a look-alike counts as a notified product. The Danish authority's SAFE tool (2021), used by authorities in about 15 countries, compares text and images against EU alert databases. Which? used Google Lens reverse image search on all 702 UK safety alerts from August 2024 to August 2025 (1,036 images). It found 798 UK-available listings "highly similar or identical" to flagged products, covering 31% of the alerts. These tools find look-alikes, but they cannot settle the question that matters for enforcement. For only 26% of Which?'s matches could analysts see the flagged hazard in the listing (a small detachable part, for example); in the rest the hazard was hidden, such as internal wiring. To check, Which? bought 15 matches (13 electrical products and 2 toys) and had independent labs test them. All 15 failed UK rules, 14 of them as unsafe, but testing every copy does not scale. Which? concludes that the current regime gives businesses "too much flexibility to determine how similar a product must be" to an alert before acting. Several of the products it tested would not have been found by searching for the identifier or brand in the alert. The platforms' own systems (eBay's AI product-safety filtering; Etsy's scanning for recalled items) are described in their transparency reports, as Which? summarises them; Which? concludes that "despite these claims, marketplaces are failing to live up to their product safety aspirations."
A new toy law (later). Regulation (EU) 2025/2509 of 26 November 2025 replaces the 2009 Toy Safety Directive. It entered into force on 1 January 2026, but under Article 59 it "shall apply from 1 August 2030"; only its chapters on notification of conformity-assessment bodies, on delegated powers and the committee procedure, and on confidentiality and penalties (Articles 28–44 and 49–55) apply from 1 January 2026. From 2030 it changes three things for marketplaces. Offers of non-conforming toys count as "illegal content" under the DSA (Article 14(1)). Marketplaces must design their pages so sellers can show the CE marking, warnings, and "the data carrier or weblink through which the digital product passport is accessible" (Article 14(3)). And the manufacturer must create a digital product passport before placing a toy on the market (Article 19), with the economic operator placing the toy on the market giving marketplaces a digital copy of the data carrier or the unique product identifier (Article 19(12)). A unique identifier could make "identical" copies easier to find. Our reading is that it does not settle look-alikes: a copy from another manufacturer would carry its own identifier, or none, and raise the same matching question. Nothing in the regulation as read defines when a different listing counts as the same toy.
What would unlock progress
An agreed, testable matching rule that turns "looks similar" into "treat as the flagged product until shown otherwise". It would combine image similarity with the features that carry the hazard: whether the part that failed (a suction cup, a rattle end, a battery cover) is visibly the same. A cheap triage step would also help: checks on a bought sample that confirm the visible hazards before an accredited lab test. Which? sets out the matching criteria its analysts used (matching identifier, identical packaging, shared design elements, only cosmetic differences). It also sets out what it threw out (batch-specific alerts, alerts with a named fix, counterfeits). These are a starting point. No test of them for agreement between raters was found. The only lab check of products they matched, in the sources read, is Which?'s own test of 15 of its matches.
How we’d know it’s being solved
- Agreement: two raters applying a written matching rule to the same set of listing–alert pairs reach a stated level of agreement (for example Cohen's kappa).
- Predictive value: among listings the rule calls a match, the share that fail the same test as the original alert when bought and tested (the independent benchmark is Which?'s 15 of 15 lab-tested look-alikes failing, only 2 of them toys; the trade bodies report TIE's 10 of 10 and, for the British Toy and Hobby Association (BTHA), 100% of the 25 look-alikes it retested, as its own page gives it and Which? reports it), and among listings it calls a non-match, the share that pass.
- Reappearance: in public data, the time from a Safety Gate alert to the last look-alike listing found, and the count of reappearances per alert, falling (the 2025 sweep counted 10 reappearances).
- A proxy to avoid: counting listings removed. Removals can rise while the same toy stays on sale under new sellers.
Where to start
Facility-free door (no partner, public data only). Take the toy alerts on the public Safety Gate portal for one six-month window. For each, search three or four marketplaces that deliver to an EU country using built-in or reverse image search. Code each result with a written matching rule that starts from Which?'s published criteria, and add hazard-feature items (is the failed part visibly the same?). Measure agreement between two coders, then refine the rule. Record how long after the alert each match is still on sale. The output is a tested matching protocol and a small open dataset of alert–listing pairs, with screenshots and dates. Skills: information design, basic statistics, careful web research. Nothing is bought.
Make door (design studio or teaching lab): a triage kit for look-alike toys. The most common failures in the toys tested were mechanical hazards: small parts, suction cups, rattle shapes, thin bags, battery covers. Design and build a low-cost bench kit and step-by-step card that lets a non-specialist confirm or rule out those hazards on a bought sample before deciding whether to send it to an accredited lab. It could include a pull jig for suction cups and small parts, a gauge set, a film-thickness check for bags, and a screwdriver-and-coin battery-cover check. What already exists: accredited labs run the full EN 71 tests. The sources read say nothing about buying the single test tools, and describe no bundled kit like this one. The kit differs by bundling only the checks for the hazards seen in these test buys, at a cost a consumer group or a local inspector could afford, with results logged against the matching rule from the first door. Safety line: the kit covers mechanical checks only, done in a supervised workshop. Chemical, electrical and magnet-strength tests stay with an accredited lab, and the step card says so; batteries stay in the toy and are not opened. The real users are a small professional group: market-surveillance and consumer-protection inspectors, and consumer groups that buy and test products. Within a term, a team can ask one regional consumer-protection or trading-standards office for a single review of the step card. The term's test uses proxy users instead: students and staff with no testing background, in a supervised bench session on campus, working on bought toys (not toys in use by children). Measure how often their kit result agrees with the published failure for toys already tested. No child takes part.
Design-the-study door. Write a protocol a consumer group or market-surveillance authority could run. It would buy a stratified sample of rule-matched and rule-unmatched look-alikes, send them to an accredited lab blind to match status, and estimate the rule's predictive value, with a pre-registered analysis plan. Access line: running it needs a buyer with a testing budget and an accredited toy laboratory; the team designs it and does not run it.
Genes
Related problems
Stuck the same way, in other fields.
field: the built worldbuilt faster than the law can read3D printed concrete walls have no structural-evaluation provisions in U.S. building codesthe rules got in the waysource: primaryfield: food & the living worlda century of saving forests built the infernoCentury of successful wildfire suppression creates catastrophic fuel loadsthe rules got in the waysource: primaryfield: health & medicinebrain data for sale, no rulesConsumer neurotechnology unregulated brain data gapthe rules got in the waysource: primaryfield: health & medicinethe app was prescribed, the outcome unknownReal-world outcome measurement for prescription digital therapeuticsthe rules got in the waysource: secondarySources
- DGCCRF, "Sécurité des articles vendus sur les marketplaces étrangères: plus de 100 000 produits retirés du marché" (Direction générale de la concurrence, de la consommation et de la répression des fraudes), 29/04/2026 (read through the Internet Archive copy of 2026-05-18, because the live page returned a bot-check page), with its annex "Résultats des prélèvements par catégorie de produits" (PDF, reached through the link on the archived page), accessed 2026-11-17. Link ↗ Link 2 ↗ Link 3 ↗
- European Commission / Consumer Safety Network, "Summary results of the first product safety sweep of the Consumer Safety Network – compliance of online marketplace offers of childcare articles with the requirements of the GPSR" (sweep of 1 April to 15 May 2025; copy hosted by Spain's Dirección General de Consumo), accessed 2026-11-17. Link ↗
- Toy Industries of Europe, "The online toy safety gap – one year on: Has anything changed" (TIE Shopping Exercise 2025 report, November 2025), accessed 2026-11-17. Link ↗
- Euroconsumers, "Systemic failures in product compliance on TEMU and Shein," 03.11.2025, accessed 2026-11-17. Link ↗
- Which? (2025), "In Plain Sight: Dangerous products for sale on online marketplaces," policy report, November 2025, accessed 2026-11-17. Link ↗
- European Commission, Directorate-General for Justice and Consumers, "The Safety Gate Rapid Alert System in 2024" (annual report; copy hosted by Hungary's consumer-protection authority), accessed 2026-11-17. Link ↗
- European Commission, Directorate-General for Justice and Consumers, "Safety Gate 2025 – Directorate-General for Justice and Consumers (DG JUST)" (web report on 2025), accessed 2026-11-17. Link ↗
- European Commission, "Product Safety Pledge + – Voluntary commitment of online platforms with respect to the safety of non-food consumer products sold online", accessed 2026-11-17. Link ↗
- Regulation (EU) 2023/988 (General Product Safety Regulation), Article 22, and Regulation (EU) 2025/2509 of the European Parliament and of the Council of 26 November 2025 on the safety of toys and repealing Directive 2009/48/EC, Articles 14, 19 and 59 (first read through the LawPlayer compiled copies; then checked against the Official Journal texts on EUR-Lex, OJ L, 12.12.2025), accessed 2026-11-17. Link ↗ Link 2 ↗ Link 3 ↗ Link 4 ↗
- Danish Safety Technology Authority, "New digital fine-toothed comb from Denmark is about to make more than 300 million European consumers safer," 1 December 2021, accessed 2026-11-17. Link ↗
- British Toy and Hobby Association (BTHA), "Update May 2025: Unsafe toys still on sale," section of its "Campaigns" page, accessed 2026-11-17. Link ↗
- European Commission (Internal Market, Industry, Entrepreneurship and SMEs), "Toy Safety in the EU", accessed 2026-11-17. Link ↗
Revision notes
This brief gets checked against its sources and contains the following revisions. Future revisions get added here.
Worth knowing
- The strongest repeat-failure result (10 of 10 look-alikes failing again) comes from the toy industry's own trade body; the independent lab test covered 15 look-alikes, only 2 of them toys.
Full check
Sources read
- DGCCRF (France), results of test buys on foreign marketplaces (2026), with its annex: full text (read in French, from an archived copy of the page)
- Consumer Safety Network, summary of the 2025 EU sweep of childcare-article offers: full text
- Toy Industries of Europe, online toy safety gap report (2025): full text
- Euroconsumers, product compliance on Temu and Shein (2025): full text
- Which?, In Plain Sight: dangerous products on online marketplaces (2025): full text
- European Commission, Safety Gate report on 2024: full text (searched for the passages used)
- European Commission, Safety Gate report on 2025 (web report): full text
- European Commission, Product Safety Pledge+ (2023): full text (including its list of signatories)
- General Product Safety Regulation (EU) 2023/988: full text (Article 22 and the relevant recitals)
- Toy Safety Regulation (EU) 2025/2509, Official Journal text: full text (the articles used, and the whole text searched)
- Danish Safety Technology Authority, SAFE tool announcement (2021): full text
- European Commission, toy-safety page: full text (the dates passage)
- British Toy and Hobby Association, campaigns page: full text (the figures passage)
What changed
- Not found in the sources, removed: in the build starting point, that the single tools used in toy safety tests are sold one by one as lab equipment; the page now says only that accredited labs run the full tests.
- Stated more strongly than the source: that Temu removed the reported items; the consumer group says Temu answered only after several weeks, saying it had removed them or was removing them.
- Contradicted by the source, corrected: that the sources do not say which marketplaces signed the voluntary safety pledge; the pledge lists its signatories, and four of the seven platforms where the failing toys were bought are on that list; that no lab test of the published matching criteria was found; Which? had 15 of its matches lab-tested, so the page now says no test of agreement between raters was found.
Left open
- the Safety Gate alert for the baby toy whose look-alikes were bought, described as the trade body reports it [no alert number given]
Separate check of specific claims: The toy trade association behind one retest result is now named in full, with the number of look-alikes it retested taken from its own page.
Separate check of specific claims: The sentence on what the EU law's 'identical content' wording leaves out is now marked as our own reading, and a related page was named.
Earlier full check: The long title was narrowed from look-alikes failing 'the same tests' to failing safety tests (display title unchanged), and over a dozen details were brought in line with the sources, including which parts of the new toy law apply early, which tool found the dangerous web addresses, and how the toy trade body and the EU sweep are described.
Separate check of specific claims: A safety line was added to the build starting point (mechanical checks only; chemical, electrical and magnet tests left to an accredited lab; batteries not opened), with who the real users are, and an unsourced claim that a standard test tool is sold cheaply was taken out.
Earlier full check: The voluntary platform pledge, the 2025 Safety Gate report, the EU product law's 'identical content' rule and the new toy law were read and added, the trade body's results now sit beside independent ones, and the long title was changed (display title unchanged).