Industries / Consumer goods / Cosmetics and personal care
Ingredients and safety testing
This is the work of showing that each ingredient and each finished product is safe to use and does what its label says, from irritation and allergy tests to checks for contaminants and measurements of sun protection.
7
Milestones achieved, from 1944 to 2025.
2
To watch for: announced, not yet confirmed outside.
3
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1944
A rabbit eye test
John H. Draize, Geoffrey Woodard and Herbert O. Calvery published methods for studying how substances applied to the skin and to mucous membranes, such as the eye, irritate and harm them. The philosopher Peter Singer wrote in 1999 that from the 1940s the eye test, known as the Draize test, was the standard test, done on fully conscious rabbits held in stocks so they could not scratch their eyes.
Credit
Draize, Woodard and Calvery
1976
An ingredient review panel
The US industry association, then the Cosmetic, Toiletry, and Fragrance Association and now the Personal Care Products Council, set up the Cosmetic Ingredient Review (CIR) with the support of the FDA and the Consumer Federation of America. An expert panel reviews the published safety data on single ingredients and states whether each is safe as used. The Council pays for it, and CIR says its panel and review process are independent of the industry.
Credit
Cosmetic, Toiletry, and Fragrance Association; FDA; Consumer Federation of America
1980
Campaign against the Draize
In 1980 the activist Henry Spira began a public campaign against the Draize test aimed at Revlon, and persuaded the company to give $750,000 over three years to a search for alternatives, according to Peter Singer's 1999 account.
LessonPublic pressure on one well-known brand moved company money into finding tests without animals.
Credit
Henry Spira; Revlon
2013
EU sales ban completed
The EU has banned animal tests of finished cosmetics since 11 September 2004 and of cosmetic ingredients since 11 March 2009, and from 2009 banned selling cosmetics tested on animals for most health effects. For three harder effects, repeated-dose toxicity, reproductive toxicity and toxicokinetics (how the body takes up and clears a substance), the sales ban began on 11 March 2013, "irrespective of the availability of alternative non-animal tests", although the Commission had reported in 2011 that such tests would not be ready. In November 2023 the EU General Court upheld a demand by the European Chemicals Agency (ECHA) that Symrise test two sunscreen filters, homosalate and 2-ethylhexyl salicylate, on animals under REACH, the EU's general chemicals law.
LessonA ban in one law does not bind another; the chemicals law can still require animal tests of an ingredient used only in cosmetics.
Credit
European Parliament and Council; European Commission
2019
Asbestos found in makeup
Talc, a soft mineral used in powders, can be contaminated with asbestos. In a year-long survey, an outside lab, AMA Analytical Services, tested 52 samples of talc-containing powder cosmetics bought by the FDA and found asbestos in 9, including products sold by Claire's and Beauty Plus Global and one lot of Johnson's Baby Powder, which Johnson & Johnson recalled on 18 October 2019. In the FDA's results table, 7 of the 9 positives found by transmission electron microscopy (TEM) had tested negative by polarized light microscopy (PLM), a cheaper method.
LessonThe test method decides the answer: light microscopy alone missed most of the asbestos found.
Credit
FDA; AMA Analytical Services
2021
Allergy hazard without animals
In 2021 the OECD, the intergovernmental body that publishes standard chemical test guidelines, issued Guideline 497. It sets fixed rules, called defined approaches, for combining results from tests that use no animals, such as a test of how a chemical binds to proteins in a test tube (Test Guideline 442C, 2015) and a test on cultured immune cells (h-CLAT, 2016), to classify whether a chemical causes skin allergy and how strongly, under the UN's Globally Harmonized System. Before then none of these tests could classify a chemical on its own, according to a 2025 history of the guidelines.
Credit
OECD member countries; validation centers including Japan's JaCVAM
2025
Sunscreens fail outside tests
The SPF on a label comes from a test in which sunscreen is applied to a panel of volunteers. Choice, an Australian consumer group, had 20 sunscreens labeled SPF 50 or 50+ tested by 10-person panels at a Sydney lab and reported in June 2025 that 16 missed their claims; Ultra Violette's Lean Screen measured SPF 4, and SPF 5 when retested at a lab in Germany. The brand's own eight retests ranged from SPF 4 to 64, and it withdrew the product. On 30 September 2025 the Therapeutic Goods Administration (TGA), Australia's regulator, said 21 products share Lean Screen's base formula, which is "unlikely to have an SPF greater than 21", and raised concerns about the reliability of PCR, the lab that had tested at least half of the sunscreens that failed Choice's test.
LessonOne lab's number on a variable test can stand on a label for years; retests at separate labs caught it.
Credit
Choice; Consumer NZ; TGA
Watch for
Australia tightens sunscreen tests
The TGA said it would push ahead with reforms under which brands must test the finished product, give the results to the TGA before a sunscreen is listed for sale and keep testing after listing, and labs testing products sold in Australia must hold international accreditation. In the same week Consumer NZ reported that six of seven sunscreens also sold in Australia missed their SPF claims.
Achieved whenthe rules take effect and a later outside test round, such as Choice's or Consumer NZ's, finds most SPF 50+ products meeting their claims.
Claimed by
TGA, October 2026
A US talc test method
US law now requires the FDA to set standard methods for detecting asbestos in talc cosmetics. The FDA proposed methods on 27 December 2024 and withdrew them on 28 November 2025, saying comments "warrant further consideration and assessment prior to issuing final regulations".
Achieved whenthe FDA publishes a final rule naming the required methods.
Claimed by
FDA, November 2025
Open
01
Failed sunscreens, retested on plates
An in vitro retest of the sunscreens at the center of the Australian disputes, using ISO 23675, the double plate method published in December 2024, in which the sunscreen is spread on UV-transparent plastic plates and a spectrophotometer measures how much UV light gets through. No product touches a person. Publish each result beside every published panel result for the same product.
The briefCredit · unclaimed02
Every talc test, one table
One public table of every published test of talc cosmetics for asbestos: product, lot, year, who tested it, the methods used (PLM, TEM, X-ray diffraction), detection limits and results, linked to the source page of each report. Then use it to show how often each method agrees with TEM.
The briefCredit · unclaimed03
Where safety panels disagree
A public table of every cosmetic ingredient reviewed by at least two of the main expert panels, the US Cosmetic Ingredient Review, the EU's Scientific Committee on Consumer Safety and Great Britain's Scientific Advisory Group on Chemical Safety (SAG-CS), with each panel's conclusion, maximum safe concentration, the effect it based that on, and date. Then list every case where the panels disagree and the reason each gives.
The briefCredit · unclaimed
Open 01 · the brief
Failed sunscreens, retested on plates
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
An in vitro retest of the sunscreens at the center of the Australian disputes, using ISO 23675, the double plate method published in December 2024, in which the sunscreen is spread on UV-transparent plastic plates and a spectrophotometer measures how much UV light gets through. No product touches a person. Publish each result beside every published panel result for the same product.
Why it is still open
Choice's and Consumer NZ's results came from tests on volunteer panels. A five-lab ring test of the double plate method, on about 32 products, found it highly reproducible, and a 2025 paper reports that it measures zinc oxide sunscreens reliably, answering an article that had doubted in vitro methods for them; neither abstract names any retail product. Regulators do not yet accept the new method as the sole support for an SPF claim, according to a 2026 review of its uptake. I found no published in vitro results for the products that failed the Choice or Consumer NZ tests.
Built with what exists
The ISO 23675:2024 method, run by a contract lab that offers it or by a university lab with its plates, robotic spreader and UV transmission spectrophotometer; Python with pandas for the comparison; Zenodo to publish spectra and reports.
First step
Build the table of every published panel result for those products, from Choice, Consumer NZ and the brands' statements, and get quotes from two labs that run ISO 23675.
Done looks like
In vitro SPF values with 95% confidence intervals for at least 20 sunscreens, including every product from the Choice 2025 and Consumer NZ 2026 tests still on sale, each from two batches; raw spectra and lab reports published; a table setting each result beside every published panel result, with the gap for each.
Open 02 · the brief
Every talc test, one table
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
One public table of every published test of talc cosmetics for asbestos: product, lot, year, who tested it, the methods used (PLM, TEM, X-ray diffraction), detection limits and results, linked to the source page of each report. Then use it to show how often each method agrees with TEM.
Why it is still open
The FDA publishes each year's results as a separate summary table, with the lab's full reports split into PDF parts of 50 to 70 MB each, and the FY19 table shows how often light microscopy and electron microscopy disagree. The FDA withdrew its proposed standard methods in November 2025, so methods still differ from lab to lab. A search of the FDA's pages and the medical literature found reviews of cancer risk but no compiled table of test results.
Built with what exists
The FDA's yearly summary tables and AMA lab reports; pdftotext from Poppler for tables with text; an AI model built for reading long documents, such as Meta's Muse Ember (see the Note "The jobs agents repeat"), to pull results from the long reports; Python with pandas; Zenodo.
First step
Turn the FDA's FY19 summary table, 52 samples, into a CSV file and count the PLM and TEM disagreements by hand.
Done looks like
Every sample in the FDA's published rounds, about 50 a year, entered with both PLM and TEM results where given; a second person re-enters a random 10% and agrees on at least 98% of fields; a published count of PLM and TEM agreement by year; the table and code under an open license.
Open 03 · the brief
Where safety panels disagree
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public table of every cosmetic ingredient reviewed by at least two of the main expert panels, the US Cosmetic Ingredient Review, the EU's Scientific Committee on Consumer Safety and Great Britain's Scientific Advisory Group on Chemical Safety (SAG-CS), with each panel's conclusion, maximum safe concentration, the effect it based that on, and date. Then list every case where the panels disagree and the reason each gives.
Why it is still open
Disagreements are reported one at a time in trade press, such as a September 2026 report that the SAG-CS reached a different conclusion from the EU on micron-sized silver while agreeing on six other ingredients. CIR keeps a searchable list of its ingredient conclusions. Searches of the medical literature and the web found no systematic comparison of the panels' conclusions.
Built with what exists
CIR's ingredient database and reports; the SCCS opinions published by the European Commission; the SAG-CS opinions published by the UK government; an AI model for long documents such as Meta's Muse Ember (see the Note "The jobs agents repeat"); ingredient names matched on INCI, the standard naming system for cosmetic ingredients; Python with pandas.
First step
Take every SCCS opinion since 2015, look up each ingredient in CIR's database, and count how many both have reviewed.
Done looks like
At least 200 ingredients with conclusions from two or more panels; a person checks a random 10% against the source reports and agrees on at least 95% of fields; a published list of disagreements with each panel's stated reason.