{"id":35915,"date":"2026-07-07T09:37:00","date_gmt":"2026-07-07T07:37:00","guid":{"rendered":"https:\/\/askme.it\/insights\/skin-ai-where-cosmetics-end-and-medical-devices-begin\/"},"modified":"2026-05-28T17:55:14","modified_gmt":"2026-05-28T15:55:14","slug":"skin-ai-where-cosmetics-end-and-medical-devices-begin","status":"publish","type":"insights","link":"https:\/\/askme.it\/en\/insights\/skin-ai-where-cosmetics-end-and-medical-devices-begin\/","title":{"rendered":"Skin AI: where cosmetics end and medical devices begin"},"content":{"rendered":"
\n

L’Or\u00e9al launched Beauty Genius at CES in Las Vegas in January 2024, the multi-category AI assistant integrating skincare, makeup and haircare. In the October 2024 beta, active on the L’Or\u00e9al Paris site, it recorded over 480,000 conversations in the first six months, with a distribution of roughly 50% makeup, more than 33% skincare and the remainder haircare. In 2025 Meta collaborated with L’Or\u00e9al to bring Beauty Genius to WhatsApp, expanding the access channel. Vichy launched ScalpConsultPro in eleven markets in 2024 and introduced Cell BioPrint, the countertop device for skin analysis in pharmacy. Perfect Corp, listed on NYSE as PERF since 2022, generated USD 69.2 million in 2025 (+14.9% YoY) on the YouCam suite that has cumulated more than 1.1 billion global downloads, and in November 2025 launched YouCam AI Beauty Agent, the conversational agent for beauty and skincare presented at CES 2026.<\/p>\n

Haut.AI, an Estonian B2B skin analysis AI startup based in Tallinn, consolidated in 2024-2025 partnerships with Beiersdorf, Ulta Beauty \u2014 launching SkinGPT in January 2025 \u2014 and NAOS, the group that owns Bioderma, for analysis across more than 150 data points with LATAM rollout in 2024 and global thereafter. In 2025 it launched Skin.Chat, the consumer-facing interface with product recommendation via Amazon. Revieve, one of the most mature platforms in the sector, has surpassed 60 million global users with verified case studies showing 79% completion rate and ROI up to +116%. The picture is clear: AI on the skin has moved out of the experimental phase and entered enterprise commodity status. The remaining constraint is regulatory.<\/p>\n<\/section>\n

\n

From selfie to conversational agent<\/h2>\n

The historical evolution of skin AI platforms can be read across three distinct generations. The first, from 2018 to 2022, was virtual try-on: the consumer’s selfie was used to simulate application of a makeup product or to measure some basic skin parameters. L’Or\u00e9al acquired ModiFace, the Canadian company that pioneered the category, on March 16, 2018 \u2014 the first tech acquisition in the group’s 109-year history, with economic terms never publicly disclosed. Vichy SkinConsultAI, active since 2019, is one of the most mature outputs of this phase.<\/p>\n

The second generation, from 2022 to 2024, was multi-parameter predictive analysis. Platforms began extracting from the consumer’s selfie parameters such as wrinkles, spots, pores, hydration and oiliness, with product recommendation based on matching the consumer’s skin profile against the brand’s catalogue. Haut.AI, Revieve, Perfect Corp and Haut.AI dominated this phase. The third generation, active since 2024-2025, is the conversational agent: the consumer no longer answers a static questionnaire but dialogues with an assistant that integrates skin analysis with clinical history, preferences, behavioural signals and dynamic recommendation. Beauty Genius, YouCam AI Beauty Agent and Skin.Chat are the most visible products in this category, and the integration of general-purpose LLMs into the workflow is the main driver of the qualitative leap.<\/p>\n<\/section>\n

\n

Clinical accuracy and the skin-tone bias<\/h2>\n

The peer-reviewed literature published in 2024 and 2025 has consolidated a picture of AI accuracy in dermatological pathology classification that is comparable to that of the specialist dermatologist in selected settings. An NCBI meta-analysis 2024-2025 confirmed comparable performance on the classification of melanoma and other skin pathologies on reference datasets. Consumer apps such as Skinive claim sensitivity of 97.4%, specificity of 93.1%, accuracy of 94.2%, but on data self-reported by the manufacturers. Aysa was evaluated in independent studies with accuracy of 59.4%. The gap between self-declared performance and independently verified performance remains a critical point in the sector.<\/p>\n

The structural constraint of all skin AI models documented by academic research is the bias toward light skin in training datasets: diagnostic accuracy decreases significantly for higher Fitzpatrick phototypes, with a risk of under-detection in populations with darker skin. For cosmetics companies launching skin AI products in markets with diversified ethnic composition \u2014 United States, United Kingdom, LATAM and African markets \u2014 the constraint is not only ethical, it is regulatory: uniform accuracy claims are not sustainable if evidence is collected predominantly on light-skinned populations. External validation on diversified datasets is the technical specification that separates a defensible product launch from an exposed one.<\/p>\n<\/section>\n

\n

The regulatory boundary: 1223\/2009 vs MDR vs AI Act<\/h2>\n

The constraint point that will determine the future of skin AI in the European market is the product’s regulatory classification. A platform that analyses the consumer’s skin and proposes cosmetic recommendations based on aesthetic preferences falls under EC Regulation 1223\/2009 on cosmetic products, with the guidelines of EU Regulation 655\/2013 on claims. A platform that diagnoses a pathology \u2014 even just a “possible suspicious spot for the dermatologist to check” \u2014 moves into the perimeter of Medical Device Regulation 2017\/745, in force since May 26, 2021 with transitional extension via Reg. EU 2024\/1860. Software with medical purpose \u2014 Software as a Medical Device, SaMD, and Medical Device Software, MDSW \u2014 requires CE marking via notified body, with classification depending on the criticality of the supported clinical decision.<\/p>\n

MDCG 2025-6, published in 2025, clarifies the interaction between MDR and the EU AI Act: SaMD devices incorporating high-risk AI must satisfy both conformity assessment procedures and the corresponding technical documentation. The leap from cosmetic to medical device changes everything: launch timelines move from a few months to one-two years, costs increase between 10x and 50x, with post-market surveillance, vigilance and audit obligations. Cosmetics companies presenting their skin AI as “pre-diagnostic analysis” or “preliminary screening” sit in the grey area that MDR and the AI Act are progressively closing. Aggressive marketing on diagnostics is becoming a real operational risk, not just a reputational one.<\/p>\n<\/section>\n

\n

Italian cases: Cosmoprof, Dermaself, CAIOME<\/h2>\n

The Italian skin AI ecosystem exists, and it has consolidated around the sector’s main trade event: Cosmoprof Bologna 2025, March 20-23, recorded over 255,000 visitors from 150 countries, with a dedicated Beauty Tech Area in Hall 14. The 2025 skincare award went to CAIOME, an Italian startup specialising in AI for cutaneous microbiome analysis and personalised treatments. Dermaself, active with a medical path that includes dermatologist Adele Sparavigna, presented its proposal for personalised AI dermo routines. Haut.AI is partner of Cosmoprof Bologna 2026 with a dedicated programme “AI Skin Intelligence” that will bring the AI integration of exhibiting brands to the centre of the event.<\/p>\n

For Italian dermo-cosmetics companies in 2026 skin AI represents a double opportunity: differentiating their pharmacy proposition through in-store consumer engagement tools and accelerating product innovation through data. The in-pharmacy skin analysis kit \u2014 Vichy’s Cell BioPrint is its most structured example \u2014 directly connects data generation to product purchase, reducing the decoupling between advisory and conversion that characterizes the pure digital channel. The Italian operational constraint is twofold: GDPR consumer consent for skin data processing, which is highly sensitive biometric data; and the regulatory classification of the platform used, which must remain below the SaMD threshold in order not to trigger MDR obligations. Italian pharmacies that integrate skin AI into their operating model in 2026 are building a significant competitive asset. Those doing so without evaluating the regulatory framework expose their channel to a non-marginal risk.<\/p>\n<\/section>\n","protected":false},"excerpt":{"rendered":"

L’Or\u00e9al Beauty Genius recorded 480,000 conversations in the first six months after launch. Perfect Corp generated USD 69 million in 2025 with YouCam. Haut.AI is in partnership with Beiersdorf, NAOS Bioderma and Ulta Beauty. When AI promises diagnostics it shifts from cosmetic to medical device, falling under MDR and the high-risk EU AI Act.<\/p>\n","protected":false},"featured_media":35917,"menu_order":0,"template":"","insights_category":[968],"insights_tags":[951,949,952,950,948],"class_list":["post-35915","insights","type-insights","status-publish","has-post-thumbnail","hentry","insights_category-ai-and-pharma","insights_tags-haut-ai","insights_tags-loreal-beauty-genius","insights_tags-mdr","insights_tags-perfect-corp","insights_tags-skin-ai"],"acf":[],"_links":{"self":[{"href":"https:\/\/askme.it\/en\/wp-json\/wp\/v2\/insights\/35915","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/askme.it\/en\/wp-json\/wp\/v2\/insights"}],"about":[{"href":"https:\/\/askme.it\/en\/wp-json\/wp\/v2\/types\/insights"}],"version-history":[{"count":1,"href":"https:\/\/askme.it\/en\/wp-json\/wp\/v2\/insights\/35915\/revisions"}],"predecessor-version":[{"id":35916,"href":"https:\/\/askme.it\/en\/wp-json\/wp\/v2\/insights\/35915\/revisions\/35916"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/askme.it\/en\/wp-json\/wp\/v2\/media\/35917"}],"wp:attachment":[{"href":"https:\/\/askme.it\/en\/wp-json\/wp\/v2\/media?parent=35915"}],"wp:term":[{"taxonomy":"insights_category","embeddable":true,"href":"https:\/\/askme.it\/en\/wp-json\/wp\/v2\/insights_category?post=35915"},{"taxonomy":"insights_tags","embeddable":true,"href":"https:\/\/askme.it\/en\/wp-json\/wp\/v2\/insights_tags?post=35915"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}