Proton Lumo 2.0 adds private multimodal AI with memory

Lumo 2.0 is a strong signal that “sovereign” and privacy‑preserving AI is maturing beyond demo status. Proton is not trying to compete with OpenAI or Anthropic on headline benchmarks so much as on trust, data jurisdiction, and predictable behavior—using open models like GLM‑5.2 and Qwen 3.5 behind a tightly controlled, European infrastructure envelope. By layering multimodality, configurable reasoning modes, and persistent memory on top of that, Lumo 2.0 closes much of the UX gap with US‑centric assistants while keeping a hard line on zero‑access encryption and no training on user data. ([macrumors.com](https://www.macrumors.com/2026/06/30/proton-lumo-version-2-new-features/))

In the race to AGI, this matters less as a frontier‑capabilities story and more as a governance and diversification story. If Europe wants meaningful leverage over the trajectory of advanced AI, it needs viable alternatives to US and Chinese stacks that enterprises and governments can adopt without triggering regulatory and political alarm bells. Lumo 2.0 is one of the first widely‑covered examples of a consumer‑grade assistant that fits that bill. It also shows how much can be done through system design—memory controls, transparent web citations, and local infrastructure—without needing to own the underlying model weights.

For the major labs, the lesson is that model quality alone is no longer the whole product: data governance, jurisdiction, and user agency over memory and context are now front‑line differentiators, especially in regions with strict privacy regimes.