Proton, a Swiss tech company known for its privacy-focused tools, has added a new feature to its Lumo AI assistant by integrating Apertus 1.5, an open-source language model developed by ETH Zurich, EPFL, and CSCS. This integration allows users to contribute to improving the model through a voluntary feedback system. If users rate a response negatively, the associated question and answer are anonymously sent to Apertus researchers for review. This feature is optional, and users can choose to switch back to other models at any time through a dropdown menu in the Lumo interface. Apertus is a fully open model, meaning its training data, weights, and methods are publicly available. It was trained on content from over 1,000 languages and respects removal requests from rights holders, as well as filters personal information. The model, launched in September 2025, has 70 billion parameters and is distributed under the Apache 2.0 license, which allows for free use and modification. However, in Lumo, Apertus 1.5 is not enabled by default, ensuring users have control over which model they use for their interactions. Conversations in Lumo are protected by zero-access encryption, meaning only the user's device can decrypt the data. If a user does not provide a rating, no information is shared with Apertus researchers. Proton notes that tens of millions of Lumo users can participate in this feedback system, though users are advised to avoid rating exchanges that contain personal information if they do not want those details to be seen by others. Apertus is designed for common tasks, while more complex reasoning is better handled by other models in Lumo, such as Qwen and GLM, which are already available in the version 2.0 tested by Proton. Imanol Schlag from the ETH AI Center described the integration as a key step toward a fully open model developed with public funds. Apertus is not exclusive to Proton; Euria, the AI assistant from Infomaniak (acquired by Proton), already offers it under the label "Swiss model in development," alongside two profiles based on Qwen 3.5. However, the protection mechanisms differ between the services, with Euria using an ephemeral mode and Lumo encrypting saved conversations. Eamonn Maguire, director of AI at Proton, emphasized the need for a European AI to remain competitive to be truly used and adopted.