Proton launches privacy-focused Lumo 2.0 with image generation and memory

Lumo 2.0 is Proton's largest upgrade to its privacy-first assistant, built around principles of no logs, zero-access encryption, no data sharing, and no training on conversations. New capabilities include multimodal image recognition and generation within a single encrypted conversation, user-controlled memory, encrypted Projects, Custom Lumos, a reasoning mode, and enhanced web search delivering live results and source citations for more accurate, transparent answers.

The encryption architecture is the differentiator: every conversation is zero-access encrypted and never logged or used for training, with Lumo for Business keeping data on independent European infrastructure that Proton says isn't subject to US Executive Orders or American data collection requests — a pointed pitch to organizations wary of US-hosted AI.

Proton says more than 10 million people have adopted Lumo as a private alternative to mainstream platforms, positioning it against ChatGPT, Gemini, and Claude on privacy rather than raw capability. The story fits the week's theme of AI-and-the-state — from OpenAI's government equity talks to Anthropic's export saga — by staking out data sovereignty as a selling point. The caveat, per corroborating coverage, is that Lumo's claimed benchmark jump remains vendor-sourced pending third-party reproduction, and privacy-first models built on open-source LLMs typically trail frontier labs on capability. Watch whether enterprise privacy demand drives meaningful adoption.