Industries / Consumer goods / Toys
Safety testing
Safety testing checks that a toy cannot choke, cut, poison or otherwise hurt a child, by measuring it against laws and standards, often in a laboratory outside the company that made it.
7
Milestones achieved, from 1979 to 2025.
1
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1979
A cylinder for small parts
A rule the US Consumer Product Safety Commission (CPSC) issued in June 1979 bans toys meant for children under three that are, or that shed, parts small enough to fit entirely inside a standard test cylinder, either as sold or after "use and abuse" tests that copy rough play. The CPSC gives the cylinder's width as 31.7 mm and notes that Europe's toy standard EN 71-1 uses a cylinder of the same size.
Credit
US Consumer Product Safety Commission
1988
One toy law for Europe
Council Directive 88/378/EEC of 3 May 1988 set essential safety requirements for every toy sold in the European Community, a toy being any product meant for play by children under 14, and member states applied it from 1 January 1990. Toys that meet the harmonized European standards are presumed to comply. The technical detail is written by the standards bodies CEN and CENELEC, and every toy sold in the EU must carry a CE marking, the maker's declaration that it meets the requirements, according to the European Commission. Directive 2009/48/EC replaced the 1988 law with stricter limits on chemicals.
Credit
Council of the European Communities
2007
Lead paint recalls at Mattel
From August to October 2007, Mattel and its Fisher-Price unit recalled toys, among them the "Sarge" die-cast car, Barbie accessories and licensed character toys, whose paint held more lead than the 0.06% by weight that US law had allowed since 1978. About 95 models were over the limit. In 2009 the companies agreed to pay a $2.3 million civil penalty, which the CPSC called the highest it had levied for a regulated product, while denying that they knowingly broke the law. The CPSC's acting chairman said the recalls helped spur Congress to pass the 2008 law below.
LessonThe lead limit had stood for 30 years; what failed was checking that each batch of toys met it.
Credit
US Consumer Product Safety Commission; Mattel; Fisher-Price
2007
Aqua Dots recalled
Aqua Dots, a Spin Master craft kit of beads that stick together when sprayed with water, was recalled on 7 November 2007 after children who swallowed beads fell ill. The beads contained 1,4-butanediol, a chemical the body turns into GHB, a sedative drug, and Spin Master had imported and sold about 750,000 units in the US. In 2011 Spin Master agreed to a $1.3 million civil penalty over CPSC staff allegations that it had not reported the illnesses in time, while denying that it knowingly broke the law.
LessonA toy maker must pass reports of harm to the regulator at once; CPSC staff said Spin Master had reports of ill children by mid-October 2007 and did not report them.
Credit
US Consumer Product Safety Commission; Spin Master
2008
Outside testing becomes law
The US Consumer Product Safety Improvement Act (CPSIA), signed on 14 August 2008, made the voluntary toy standard ASTM F963 mandatory, cut the lead limit in paint from 0.06% to 0.009%, and required children's products covered by safety rules to be tested by an outside laboratory accredited for the purpose, not only by their makers. It also extended the choking warnings required on toy packaging to Internet advertising, from 12 December 2008.
Credit
US Congress; US Consumer Product Safety Commission
2025
Limits on water beads
Water beads are small balls of absorbent polymer that can swell to up to eight times their dry width in water; swallowed, they can block a child's gut. The CPSC estimates that 6,300 injuries linked to water beads were treated in US emergency departments from 2017 to 2022. Its rule, published on 12 December 2025 and in force from 12 March 2026, requires water beads in toys to pass through a 5.0 mm gauge when fully swollen, limits the acrylamide, a toxic chemical left over from making the polymer, that can be drawn out of them, and requires warning labels.
Credit
US Consumer Product Safety Commission
2025
Marketplace toys fail tests
Toy Industries of Europe (TIE), the trade association of European toy makers, bought 70 unbranded or little-known toys from sellers outside the EU on seven online marketplaces, among them AliExpress, Amazon Marketplace, Shein and Temu, and sent them to an accredited outside laboratory. 96% did not comply with EU rules and 86% had serious safety problems, most often small parts that a young child could choke on. In the US, the Toy Association reported in November 2025 that more than 70% of 29 toys bought from Temu and Shein failed at least one test at a government-accredited laboratory.
LessonTIE found that most toys failed on design, not on variation between samples, so basic laboratory testing would have caught every problem.
Credit
Toy Industries of Europe; The Toy Association
Watch for
A digital passport per toy
The EU's Toy Safety Regulation, Regulation (EU) 2025/2509, adopted on 26 November 2025, requires every toy sold in the EU to have a digital product passport: its compliance information, reached through a data carrier on the toy, its packaging or its papers. Customs authorities are to check automatically that imported toys have a valid passport reference. The regulation also bans the intentional use of PFAS, the long-lasting "forever chemicals", in toys, along with a listed group of bisphenols. It entered into force on 1 January 2026 and applies from 1 August 2030.
Achieved whenafter 1 August 2030 the EU passport registry holds entries for toys on sale and customs authorities report stopping toys at the border for lacking one.
Claimed by
European Parliament and Council of the EU, November 2025
Open
01
Choking-warning audit of listings
A repeatable public audit of toy listings on large online marketplaces that checks whether listings for toys with small parts, small balls, marbles or latex balloons show the choking warning that US law has required in Internet advertising since December 2008. A browser agent collects public listing pages at a slow rate, a model decides whether each product needs a warning and whether the page shows one, and people check a random sample.
The briefCredit · unclaimed02
Age grading from listings
A model that reads a toy listing's photos and text and assigns the age group the toy suits, following the CPSC's 2020 Age Determination Guidelines, and flags listings whose stated age is older than the toy's features suggest. The age grade decides which tests apply: a toy for a child under three must have no small parts.
The briefCredit · unclaimed
Open 01 · the brief
Choking-warning audit of listings
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A repeatable public audit of toy listings on large online marketplaces that checks whether listings for toys with small parts, small balls, marbles or latex balloons show the choking warning that US law has required in Internet advertising since December 2008. A browser agent collects public listing pages at a slow rate, a model decides whether each product needs a warning and whether the page shows one, and people check a random sample.
Why it is still open
US rules require these cautionary statements on packaging and on websites that sell the toy directly. Checks so far are one-off and small: a 2018 PIRG survey found 87% of latex balloons on Amazon lacked the required warning, and the Toy Association's November 2025 report listed, toy by toy, sales pages on Temu and Shein missing the federal small parts warning. The CPSC lists "Online Marketplace Monitoring", AI that finds recalled and banned products on online marketplaces, among its AI uses for 2026; it is an internal enforcement tool, it looks for recalled and banned goods rather than missing warnings, and it publishes no rates. Web searches, GitHub and Hugging Face found no repeated, published measure across marketplaces and no open tool for it.
Built with what exists
A browser agent such as Browser Use (see the Note "Deleting 80% to start over"), a vision-language model to read listing photos and text, a model that answers many fixed-answer questions per listing such as Gemini Decision 1 (see the Note "A thousand questions, one pass"), Python, and Label Studio for the human checks.
First step
Check 100 balloon and marble listings on one marketplace by hand against the warning texts in 16 CFR 1500.19, and turn what you learn into a labeling guide.
Done looks like
2,000 toy listings from at least four marketplaces, each labeled as needing a warning or not and showing one or not, with the model's labels checked by two people on a random 300; published rates per marketplace with 95% confidence intervals, repeated three months later.
Open 02 · the brief
Age grading from listings
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A model that reads a toy listing's photos and text and assigns the age group the toy suits, following the CPSC's 2020 Age Determination Guidelines, and flags listings whose stated age is older than the toy's features suggest. The age grade decides which tests apply: a toy for a child under three must have no small parts.
Why it is still open
Age grading is done by trained people at toy makers and laboratories. The CPSC published its 2020 guidelines, which match a product's features to the skills, play and interests of children at each age, on 27 January 2020 and began using them on 1 June 2020. The Toy Association's November 2025 tests found toys advertised for children from 12 months that failed the small parts test. Web searches found no automated age grader and no published accuracy test of one.
Built with what exists
The 2020 guidelines as the marking scheme, a vision-language model such as Claude or Gemini, Label Studio, and listings from toy makers whose packaging states a lab-assigned age grade.
First step
Turn the guidelines' toy characteristics for children under three into a checklist and grade 50 listings by hand.
Done looks like
On 500 toy listings graded independently by two people trained on the guidelines, the model agrees with their agreed grade at least as often as the two people agree with each other, reported with the full table of agreements and misses.