AI: the CNIL updates its traceability tool for open-weights AI models
26 August 2026
The CNIL releases a new version of its demonstrator for exploring the genealogy of open-weights AI models. This update notably improves the tool's performance and usability, and automates data updates. An English version is now available.
Exploring the genealogy of AI models
Open-weight AI models can be downloaded, modified, fine-tuned with new data, or combined with other models before being released again. A single model (such as Kimi K3, Mistral Medium, LLaMa, etc.) can thus give rise to numerous derivative models.
The Genmod demonstrator, developed by the CNIL’s AI department in collaboration with the CNIL’s Digital Innovation Laboratory (LINC) and initially released in November 2025, allows users to explore these connections and trace a model’s ancestors (the models from which it originated) as well as its descendants (the models it contributed to).
This traceability is particularly useful for studying the consequences of AI models memorizing training data. It makes it possible to identify, starting from a model that has memorized personal data, other models in its "genealogy" that may also have retained this information. In particular, this facilitates the study of how rights under the GDPR can be exercised.
A new version that is faster and easier to use
This new version introduces several improvements to the demonstrator.
An interface available in French and English
The entire application can now be used in both languages (via a button in the top-right corner of the window).
Faster searches
The graph exploration engine has been optimized to significantly reduce the time required for extensive searches. Consequently, a search with no depth limit now takes around twenty seconds on average.
Better visibility of search progress
The interface now displays search progress and, where estimable, the time remaining until completion. Multiple searches can also be processed simultaneously. During periods of high demand, users can see their position in the queue and the number of active searches.
Regularly updated data
A new process allows the demonstrator’s underlying database to be rebuilt using public data on models and datasets available on HuggingFace. Notably, this enables the weekly automation of graph updates, making it easier to track the rapid evolution of the open-weights model ecosystem. The data update date is now displayed within the application.
Improved navigation and results viewing
The most downloaded models are now prioritized in search suggestions, and detailed results are sorted by download count by default. Direct access to the home page has also been added to the application's various pages.
Finally, the layout has been revised to ensure consistent display across major modern browsers (Firefox, Chrome, Edge, and Safari) and various screen sizes.
This new version continues the CNIL’s experiment regarding the traceability of open-weights AI models, while making the exploration of their genealogies faster and easier to update.
