BTY.01
Cosmetic Claims Compliance
Every cosmetic claim inventoried and linked to its evidence under Regulation 655/2013, and the product information file assembled with gaps flagged. The Responsible Person decides.
Open pageIndustry
Beauty AI,
with the evidence attached.
The safety assessor still signs. Claims, product information files, shade matching and formulation research, built around EU Cosmetics Regulation 1223/2009. Every claim carries its evidence, and the safety assessor and Responsible Person still sign.
Last reviewed 30 October 2026 · Engineering guidance, not legal or regulatory advice.
In context
AI in the beauty industry means using models to check marketing claims against their evidence, assemble product information files, match shades from a photo and search ingredient and formulation data. Under EU Regulation 1223/2009 a named safety assessor and Responsible Person stay accountable for every product.
Where it breaks today
01
Claims outrun their evidence
Marketing writes faster than regulatory can review. A single launch produces hundreds of claims across packs, listings, ads and influencer briefs, and each one has to meet the common criteria of Regulation (EU) 655/2013 with evidence on file.
02
The PIF is assembled by hand
The product information file pulls from suppliers, labs and the formulator: specifications, test reports, the safety report, GMP statements. It lives in folders and inboxes, and it is rebuilt every time a raw material changes.
03
Shade matching loses customers
Buying foundation online is a guess about shade and undertone, and a wrong guess comes back as a return or a lost customer. Try-on tools help, but a match that works on light skin and fails on deep skin is worse than none.
OUR APPROACH
AI for cosmetics, built around the regulation
Most writing about AI in beauty is about the shopper: skin analysis apps, AI skincare routines, chat-based product recommendations. Those are real, but the work that costs a beauty brand the most time sits behind the counter, in the regulatory file, the claims review and the lab. That is where we build.
The constraint is the EU Cosmetics Regulation (EC) 1223/2009 and its UK equivalent. Every product needs a Responsible Person, a product information file, a cosmetic product safety report signed by a qualified safety assessor, and notification in the CPNP or the UK portal. Claims must meet the criteria in Regulation (EU) 655/2013. In the US, MoCRA has added registration, listing and safety substantiation duties. A model can do the reading, cross-checking and drafting; it cannot hold any of those roles.
For consumer-facing tools such as virtual try-on and AI skin analysis, the constraint is different: face images are personal data under GDPR, and the EU AI Act prohibits using biometric data to infer sensitive characteristics. We design shade matching to work on measured skin tone and undertone, never on inferred ethnicity.
HOW IT DEPLOYS
PHASE 01
Map the file
We trace one product from formula to shelf: which documents exist, who owns them, where claims are approved and where evidence is stored.
PHASE 02
Link claims to evidence
Claims, test reports and ingredient data are structured so every statement can be traced to the study or specification behind it.
PHASE 03
Draft, then sign
The system drafts and flags; the regulatory lead, safety assessor and Responsible Person review and sign, and every decision is logged.
What we build
01
Claims review and PIF assembly
Every claim checked against its evidence and the six common criteria, and the product information file assembled from supplier and lab documents with gaps flagged.
SEE CLAIMS COMPLIANCE →02
Virtual try-on and shade matching
Shade recommendations tested across the full range of skin tones before launch, with face images processed for the session and not kept.
SEE VIRTUAL TRY-ON →03
Formulation research
Search across your formula history, supplier data and the restricted-substance annexes, so a chemist starts from evidence rather than a blank sheet.
SEE FORMULATION →04
Beauty advisor agents
Product and routine advice grounded in your catalogue and ingredient lists, that refers anything medical, such as a reaction or a skin condition, to a person.
SEE SUPPORT AGENTS →Controls we implement
EVIDENCE LINKED TO EVERY CLAIM
SIGN-OFF BY A NAMED ASSESSOR
FACE IMAGES NOT RETAINED
NO TRAINING ON YOUR FORMULAS
VERSIONED PIF RECORDS
AUDIT LOG ON EVERY APPROVAL
These are engineering controls we build into the architecture, not certifications. QQuantum.ai holds no security certifications, and the regulatory determinations and audits your products require remain yours — see our trust centre.
Beauty use cases
-
-
BTY.02
Virtual Try-On
Virtual try-on and AI shade matching tested per skin-tone band and gated on the weakest, with lighting checks, session-only face images and no inference of ethnicity.
Open page -
BTY.03
Cosmetic Formulation
AI formulation research over your formula history and supplier data, screening candidates against Annexes II to VI with citations. Testing and safety assessment stay with people.
Open page
Questions
Questions about Beauty & Cosmetics.
How is AI used in the beauty industry?
On the consumer side, for virtual try-on, shade matching, AI skin analysis and product recommendations. Behind the scenes, for checking claims, assembling regulatory files, searching ingredient and formulation data, and producing marketing content. The behind-the-scenes uses carry less publicity and more of the cost, and they are where a person’s signature is legally required, so the system has to support that sign-off rather than replace it.
Can AI check cosmetic claims for compliance?
It can do the first pass. A model can read every claim across packs, listings and ads, match each to the evidence on file, and flag those that look unsupported or that breach the common criteria in Regulation (EU) 655/2013. Deciding whether a claim is acceptable remains with your regulatory team and Responsible Person; the system makes that review faster and better documented.
Is AI skin analysis accurate?
It depends heavily on lighting, camera and how the model was tested. Many tools perform worse on deeper skin tones because their training data under-represents them. Any skin analysis we build is tested across the full range of tones you sell to before launch, and it describes visible characteristics only; it does not diagnose skin conditions, which would move it toward medical-device territory.
What does the EU AI Act mean for beauty brands?
For most beauty uses, transparency and data duties rather than heavy obligations. Chat assistants must make clear they are automated, synthetic imagery may need disclosure, and using biometric data to infer sensitive characteristics such as ethnicity is prohibited. Face images remain personal data under GDPR throughout. How these apply to your product is your determination; we design so the defaults are conservative.
Can AI formulate cosmetics?
It can suggest and screen, not formulate on its own. Models can search past formulas, propose ingredient combinations for a target texture or claim, and check candidates against the restricted and prohibited annexes of Regulation 1223/2009. Stability, preservative efficacy and safety are still established at the bench and signed by a qualified safety assessor.
