By Melike Sönmez, Head of Product at EML Techs
The EU’s longer list of labelled fragrance allergens is a useful test of what automated label review can and cannot do.
The EU has expanded the list of fragrance allergens that must be named on cosmetic labels from 26 to 81 (2). New products have had to comply since the end of July 2026, and products already on sale have until mid-2028 (5).
For a brand with dozens of fragranced products, that means rechecking every formula, collecting new data from fragrance suppliers and updating every ingredient list, often in several languages. None of it is difficult. It is large and repetitive, which is where software can genuinely help, and also where it is easy to expect too much from it.
What changed, and who it affects
The update amends the annex of the EU Cosmetics Regulation that lists restricted substances (1, 2). Allergens above a small threshold must be named individually on the label, and the threshold is lower for leave-on products than for rinse-off ones (5). Some substances were added, some entries were widened, and a couple moved to the banned list.
Detergents have their own rules, which follow the cosmetics list. Today, Regulation (EC) No 648/2004 requires allergenic fragrances added above 0.01% by weight to be listed on a detergent label, using the allergen list from cosmetics legislation (3). Its replacement, Regulation (EU) 2026/405, entered into force in March 2026 but only applies from 23 September 2029 (4). Until then the current requirements stand. The new regulation keeps fragrance allergen disclosure aligned with the cosmetics rules, and products placed on the market under the old rules before the switch can stay on sale (6). Brands that use the same fragrance in personal care and household products can work from one up-to-date supplier declaration for both.
A practical approach
Most of an allergen review is cross-referencing, and it breaks down into four steps.
- Start from data, not the label. The fragrance supplier’s allergen declaration says how much of each allergen the fragrance contains. Combined with the fragrance dose, it shows which allergens cross the threshold in the finished product.
- Compare with the artwork. Tools that read label artwork can extract the printed ingredient list, including from curved and multilingual packs, and flag allergens that are missing, misspelled or named incorrectly.
- Keep the rules current. Reference data must change when the law changes. A useful tool shows the rule behind each finding, so a reviewer can check it quickly.
- Re-run on every change. A new supplier, a reformulation or a new market language each triggers another round. The first check is rarely the expensive one.
A worked example, with illustrative figures, shows how the first step works in practice. A fragrance supplier declares that its fragrance oil contains 2% linalool, 0.5% limonene and 0.05% geraniol. The level of each allergen in the finished product is its share of the fragrance multiplied by the fragrance dose. That level is then compared with the threshold: 0.001% for leave-on products and 0.01% for rinse-off products (5).
| Allergen (share of fragrance) | Hand cream: leave-on, 0.5% fragrance | Shower gel: rinse-off, 1% fragrance |
| Linalool (2%) | 0.01%, must be labelled | 0.02%, must be labelled |
| Limonene (0.5%) | 0.0025%, must be labelled | 0.005%, not required |
| Geraniol (0.05%) | 0.00025%, not required | 0.0005%, not required |
The same fragrance gives two different labels. The hand cream must name linalool and limonene. The shower gel names only linalool, even though it contains twice as much fragrance. If an allergen also comes from another ingredient, such as an essential oil, the amounts are added together before the comparison. Copying an ingredient list from one product to the next would get one of these labels wrong.
AI is especially useful for the reading and matching: pulling text from artwork in many languages, recognising name variants and drafting a corrected ingredient list for a person to approve.
What automation cannot do
- It is only as good as its inputs. No tool can work out allergen content from the word “Parfum”. If supplier data is missing or out of date, the honest output is a flag, not a guess.
- General chatbots are not reference sources. They answer from training data that may predate the change, and they can cite rules that don’t exist. Anything used for compliance should point to a source a person can open and check.
- Judgement stays human. Whether a claim is substantiated, how a borderline product is classified, how a transition period applies to a specific product: these need someone who knows the product and the rule.
- Responsibility doesn’t move. Under EU law, the Responsible Person answers for compliance and a qualified assessor carries out the safety assessment (1). Software can prepare and check. People decide and sign.
Used this way, AI is a fast first reader. It handles the volume, catches the obvious misses and shows its sources, leaving experts free for the questions that need a person. With rules now changing several times a year, that division of work is likely to matter more, not less.
References
- Regulation (EC) No 1223/2009 of 30 November 2009 on cosmetic products, EUR-Lex.
- Commission Regulation (EU) 2023/1545 of 26 July 2023 amending Regulation (EC) No 1223/2009 as regards labelling of fragrance allergens in cosmetic products, EUR-Lex.
- Regulation (EC) No 648/2004 of 31 March 2004 on detergents, EUR-Lex.
- Regulation (EU) 2026/405 of 11 February 2026 on detergents and surfactants, EUR-Lex.
- Compliance countdown: EU fragrance allergens expanded to 81 substances, CIRS Group.
- EU updates Regulation (EU) 2026/405 on detergents and surfactants, SGS, April 2026.
About the author: Melike Sönmez is Head of Product at EML Techs, which develops Export AI, software that checks cosmetic and household product labels against EU and UK rules. She previously worked in supply chain and product roles at Unilever, P&G and Amazon.
