A beauty assistant inside ChatGPT

At VivaTech 2026, L’Oréal announced the integration of Maybelline’s virtual make-up try-on into ChatGPT. It’s not an external plugin: the feature leverages ModiFace technology, developed by the Canadian company acquired in 2018, and lets users test make-up looks through a conversational interface. Users describe the style they want or upload a photo, and the system generates a realistic preview. L’Oréal’s Beauty Tech services already surpassed 120 million uses across 66 countries and 31 brands by the end of 2025, a sign of adoption beyond the niche.

From storefront to research: GPT‑Rosalind and the microbiome

The collaboration goes beyond the consumer. L’Oréal is using GPT‑Rosalind, OpenAI’s life‑sciences reasoning model, to map the skin microbiome. Starting with La Roche‑Posay, the goal is to identify beneficial bacteria that could inspire new skincare products. The company is already working with IBM on a Formulation Foundation Model and with NVIDIA on 3D rendering and predictive formulation: generative AI becomes an engine of scientific innovation, not a gimmick.

Advertising and product discovery: the commercial front

In the United States, L’Oréal will improve the visibility of brands like Lancôme and Kérastase directly within ChatGPT conversations. The global advertising pilot involves SkinCeuticals, CeraVe, and Garnier, with native ads placed at moments of purchase intent. The group’s e‑commerce exceeded 30% of sales in 2025, growing double digits: the bet is that “AI‑assisted” interaction accelerates conversion, though details on ad placement remain unclear.

The internal creative engine: CreAItech

Internally, OpenAI’s latest model will power CreAItech, the generative platform that creates images and videos while respecting each brand’s visual identity. Over 73,000 employees have already been trained in generative AI, with tools such as L’OréalGPT and personal AI companions. The goal is to speed up beauty content production without diluting creative heritage.

AI‑RADAR: cloud speed vs. on‑premise control

L’Oréal’s deal with OpenAI illustrates how a cosmetics giant fully embraces cloud infrastructure to innovate on all fronts. But the choice reignites the data‑sovereignty debate. Facial scans for virtual try‑on and microbiome data are sensitive personal information: in Europe, GDPR imposes strict constraints. Entrusting processing and storage to a third‑party provider, however certified, raises questions about transparency and control. For IT decision‑makers who follow AI‑RADAR, the L’Oréal case is a concrete example of the trade‑off between time‑to‑market and technological autonomy. The alternative exists: running open‑weight models on on‑premise GPU clusters keeps data within corporate boundaries, but demands hardware investment, MLOps expertise, and a careful TCO assessment. Today, convenience tips the scale, but as regulations evolve and local stacks mature, we will likely see a rebalancing.