Proton releases Lumo 2.0, a private AI chat that does not train on user data

In July 2026, Proton released Lumo 2.0, the second generation of its in-house AI chatbot, pitched as a private alternative to ChatGPT. The product's central promise: user conversations never enter training datasets, and the entire infrastructure complies with European data protection law.

How Proton protects data in Lumo 2.0

Proton has spent more than a decade building a reputation on technically grounded privacy: ProtonMail cannot see the contents of emails, and ProtonVPN keeps no connection logs. Lumo 2.0 follows the same architectural logic: chatbot conversations are stored encrypted, which rules out access to their contents, including by Proton's own employees.

Key features of the service:

• Conversations are not used for training, neither for the current model version nor for future ones
• End-to-end encryption of conversations, similar to ProtonMail
• Server infrastructure is located in Switzerland, one of the strictest jurisdictions for privacy law
• Full GDPR compliance, the European regulation on personal data protection
• Lumo 2.0 is the second generation of Proton's chatbot

Who needs Lumo 2.0, and why now?

Most popular AI assistants, including ChatGPT, Gemini and Microsoft Copilot, may use user conversations by default to improve their models unless the user explicitly opts out. Even with training switched off, conversations are stored on the servers of American companies, which creates legal uncertainty for organizations operating under European regulation.

The real audience for Lumo 2.0 is journalists, lawyers, medical professionals, HR specialists and corporate teams under strict industry oversight. For them, a leak of work conversations through a training dataset means legal liability or reputational damage. In this context, a technically grounded promise of "we physically cannot read your conversations" is worth more than any performance benchmark.

Proton's approach implements the principle of privacy by design: privacy is built into the product's architecture rather than being a setting the user has to enable. This principle underlies GDPR, and Proton applies it consistently across every product in its ecosystem.

How does Lumo differ from other private chatbots?

The market already offers several options for people who do not trust the large cloud platforms. Local tools such as Ollama and LM Studio run models directly on the user's device: this gives maximum control but requires powerful hardware and technical skill. Cloud services that claim to be private offer convenience but fall short on brand trust.

Lumo 2.0 sits between the two extremes: a familiar cloud UX plus a reputation backed by years of operation without leaks and without handing over data at the first request from authorities. Proton is not an AI lab. By all appearances, the company uses a third-party language model with a private infrastructure layer added on top. The company does not disclose which LLM runs under Lumo's hood.

What this means

Lumo 2.0 does not compete with ChatGPT on model capability; it competes on trust. For Proton, launching a chatbot is a logical extension of its ecosystem: a product built on the same values that made ProtonMail the main Gmail alternative for privacy-conscious professionals. As AI tools become more deeply embedded in workflows, demand for "private AI" will only grow.