Apertus 1.5 Joins Proton Lumo for Swiss Open AI Feedback Loop
ETH Zurich, EPFL, and CSCS researchers announced on September 17, 2026 that privacy company Proton is integrating Apertus 1.5 into Lumo, Proton’s Swiss-built AI assistant, making the fully open model selectable for tens of millions of users. Users who opt in can send anonymised feedback on Apertus answers back to the academic team—something open university models usually lack—while Lumo chats remain protected by zero-access encryption.
Filed under Product and dated September 17, 2026, this AI4Switzerland briefing treats the partnership as Swiss open-sovereign AI news rather than a closed frontier launch. Apertus 1.5 is trained on CSCS’s Alps supercomputer, released under Apache 2.0 with documented data, code, and development, and already surfaces through Swisscom, Infomaniak, OnPrem.ai, and other partners. Co-leads Imanol Schlag and Martin Jaggi call real-world voluntary feedback a turning point for improving usefulness and robustness without surrendering privacy defaults.
Why it matters: Swiss and European buyers want competitive open models that still learn from usage. A Lumo feedback loop can accelerate quality—but only if consent is genuine, anonymisation holds, and opt-outs stay easy.
What it means in practice
Swiss product, public-sector, and enterprise leads should test Apertus inside Lumo on local-language tasks; confirm feedback toggles and encryption claims; assign an owner for model-version pinning; run time-boxed comparisons against proprietary assistants on citation quality; and prefer deployments that keep logs on Swiss soil where policy requires. Place the launch beside Apertus on-demand Alps service and CLUE learning-sciences centre.
Caveats come first. Optional feedback volumes can stay thin; open weights do not equal enterprise SLAs; and encryption marketing still needs independent audits. AI4Switzerland therefore presents the Proton integration as directional product context until published feedback volumes and quality deltas appear.
What to watch next: default model rankings inside Lumo; first published Apertus improvement cycles from feedback; and how Swiss AI Weeks outreach drives consumer trials. Readers can continue on the AI4Switzerland homepage, or browse the Newsroom for additional briefings.
Bottom line: treat this update as orientation, not instruction. Swiss open AI is entering mass assistants and still early. Organizations that benefit most will verify consent controls, measure answer quality honestly, and refuse to confuse a partnership notice with finished commercial parity.